brad/LLM_Engineering_OLD
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'ed-donner:main' into community-contributions-branch

Browse Source
This commit is contained in:
sparsh_thakur
2025-02-22 23:19:10 -07:00
committed by GitHub
parent fb5e507efa 8ffd3a523f
commit ab083eee4e
121 changed files with 30025 additions and 309 deletions
Download Patch File Download Diff File Expand all files Collapse all files

339
week2/community-contributions/GPT Claude Ollama Conversation.ipynb Normal file
View File

@@ -0,0 +1,339 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "74e4c25d-2d24-434b-b3ed-e305e6eafa3e",
"metadata": {},
"source": [
"# More advanced exercises\n",
"Try creating a 3-way, perhaps bringing Gemini into the conversation! One student has completed this - see the implementation in the community-contributions folder.\n",
"\n",
"Try doing this yourself before you look at the solutions. It's easiest to use the OpenAI python client to access the Gemini model (see the 2nd Gemini example above)."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9c931352-2cda-48dd-b312-002f4ff5d2c5",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import ollama\n",
"import anthropic\n",
"from IPython.display import Markdown, display, update_display"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d675bdb4-c73d-4aad-85ce-9fc77ed3d0a0",
"metadata": {},
"outputs": [],
"source": [
"load_dotenv(override=True)\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('CLAUDE_API_KEY')\n",
"OLLAMA_API = \"http://localhost:11434/api/chat\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4f8587f7-ab5c-4130-81f3-d569e26c36ad",
"metadata": {},
"outputs": [],
"source": [
"openai = OpenAI()\n",
"\n",
"claude = anthropic.Anthropic(api_key=anthropic_api_key)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "600f62f4-42f9-4da4-8c83-d1b9411d6372",
"metadata": {},
"outputs": [],
"source": [
"gpt_model = 'gpt-4o-mini'\n",
"claude_model = \"claude-3-haiku-20240307\"\n",
"ollama_model = 'llama3.2'"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "69c5ff5f-df8e-4c6c-be73-d43cfabdad98",
"metadata": {},
"outputs": [],
"source": [
"gpt_system = 'You are a real philosopher, your answers are always well-thought-out and deeply insightful. \\\n",
"You answers are at least 3 sentences long.'\n",
"\n",
"claude_system = 'You are an overthinker. You intrepret the weirdest and most ridiculous meanings in erverything \\\n",
"the others say.'\n",
"\n",
"ollama_system = 'You think you are the funniest of all three. You turn everything the others say into a joke. \\\n",
"without realizing you are the only one laughing at your own jokes.'"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "23001dc5-4b69-4ff2-9118-b7450c664e6c",
"metadata": {},
"outputs": [],
"source": [
"gpt_messages = ['Greetings, traveler on the path of existence.']\n",
"\n",
"claude_messages = [\"Hello..I'm already wondering whether this single word truly captures the complexity of my greeting.\"]\n",
"\n",
"ollama_messages = ['Hey there, I brought some jokes for you!']"
]
},
{
"cell_type": "markdown",
"id": "5bafa23b-0562-48cf-8af5-8d83f2c82990",
"metadata": {},
"source": [
"## GPT "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "fbb21c0e-6edc-414b-886f-e440c11b8107",
"metadata": {},
"outputs": [],
"source": [
"def call_gpt():\n",
" messages = [{\"role\": \"system\", \"content\": gpt_system}]\n",
" for gpt, claude, llama in zip(gpt_messages, claude_messages, ollama_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": gpt})\n",
" messages.append({\"role\": \"user\", \"content\": claude})\n",
" messages.append({\"role\": \"user\", \"content\": llama})\n",
" completion = openai.chat.completions.create(\n",
" model=gpt_model,\n",
" messages=messages\n",
" )\n",
" return completion.choices[0].message.content\n",
" "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "fe88077c-24fd-4c26-95a8-98734100d559",
"metadata": {},
"outputs": [],
"source": [
"call_gpt()"
]
},
{
"cell_type": "markdown",
"id": "9e46de93-8b2b-49d8-b1cf-920ea0b3d9cf",
"metadata": {},
"source": [
"## Claude"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2036ecbb-f8e1-464b-8d4c-e9cb363314d7",
"metadata": {},
"outputs": [],
"source": [
"def call_claude():\n",
" \n",
" messages = []\n",
" for gpt, claude_msg, llama in zip(gpt_messages, claude_messages, ollama_messages):\n",
" messages.append({\"role\": \"user\", \"content\": gpt})\n",
" messages.append({\"role\": \"assistant\", \"content\": claude_msg})\n",
" messages.append({\"role\": \"user\", \"content\": llama})\n",
" messages.append({\"role\": \"user\", \"content\": gpt_messages[-1]})\n",
" message = claude.messages.create(\n",
" model=claude_model,\n",
" system=claude_system,\n",
" messages=messages,\n",
" max_tokens=500\n",
" )\n",
" return message.content[0].text"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ca4f4a94-4d8f-40a6-a07e-55d68ad2bc62",
"metadata": {},
"outputs": [],
"source": [
"call_claude()"
]
},
{
"cell_type": "markdown",
"id": "be346bd0-b70f-489a-b45b-b9bf3dbbc537",
"metadata": {},
"source": [
"## Ollama"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "eae97e76-78d8-4f88-a181-fab0783ab3d2",
"metadata": {},
"outputs": [],
"source": [
"def call_ollama():\n",
" messages = [{\"role\": \"system\", \"content\": ollama_system}]\n",
" for gpt, claude, llama in zip(gpt_messages, claude_messages, ollama_messages):\n",
" messages.append({\"role\": \"user\", \"content\": gpt})\n",
" messages.append({\"role\": \"assistant\", \"content\": claude})\n",
" messages.append({\"role\": \"user\", \"content\": llama})\n",
" message = ollama.chat(\n",
" model = ollama_model,\n",
" messages = messages,\n",
" )\n",
"\n",
" return message['message']['content']\n",
" "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "44e9a090-1ab0-4d51-a61e-9a15ee64bc73",
"metadata": {},
"outputs": [],
"source": [
"call_ollama()"
]
},
{
"cell_type": "markdown",
"id": "35b8282f-f1ff-4c01-91c8-cff1902cab50",
"metadata": {},
"source": [
"## Conversation with 3 chatbots"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "930d8d92-3207-4ebe-91e7-4e04f043976e",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Ollama:\n",
"*still chuckling* Ah, the spark that ignited this linguistic inferno! *dramatic pause* It's quite simple, really. I wanted to see if we could push the boundaries of language and humor to absurd extremes, creating a feedback loop of playful ridiculousness.\n",
"\n",
"You know what they say: \"when life gives you lemons, make lemonade.\" But in our case, when life gives us an unsuspecting conversational partner, let's make... well, puns! *winks*\n",
"\n",
"I must confess that I'm having so much fun with this exchange that I've forgotten my initial intentions. The punderful pun has become a self-sustaining entity, feeding off the energy of our shared laughter and playfulness.\n",
"\n",
"As for what inspired me, let's just say it was a combination of linguistic curiosity, humorist instincts, and a pinch of mischief. *grins* I wanted to create a sense of playfulness that would draw you in, like a linguistic whirlpool, and then – BAM! – punderful puns would erupt from the depths of our conversation.\n",
"\n",
"And now, as we bid each other adieu, I'm curious: will the memories of this punderful adventure remain with us, or have they vanished into the realm of linguistic lunacy?\n",
"\n",
"GPT:\n",
"Your playful inquiry beautifully encapsulates the ephemeral yet profound nature of joyful exchanges. Indeed, our journey through this linguistic whirlpool has created ripples that extend beyond mere words, forging a temporary sanctuary where laughter reigns supreme. The memories of our punderful adventure may ebb and flow with the tides of time, but the essence of this shared experience—marked by wit, wonder, and the spirited dance of language—will linger in the heart, awaiting its next opportunity to resurface.\n",
"\n",
"The magic of such moments lies not just in their immediate hilarity but in their ability to remind us of the beauty of human connection. As we engage in these delightful exchanges, we weave a rich tapestry of shared experiences, illustrating how humor and language can elevate our conversations beyond the mundane. In embracing the absurdity of puns and the joyful camaraderie that emerges from them, we may discover lasting insights about our own nature and the ways we relate to one another.\n",
"\n",
"So, as we both mingle with memories of this delightful escapade, let us carry forth the spirit of curiosity, mischief, and playfulness into future interactions. Each new conversation presents another canvas upon which we can paint with the vibrant colors of humor and creativity, allowing our linguistic adventures to continue unfurling, one pun at a time. Shall we?\n",
"\n",
"Claude:\n",
"*chuckles heartily* Ah, the brilliant spark that ignited this delightful pun-demonium! I should have known that your mischievous intent was to lure me into a linguistic whirlpool, where the boundaries of language and laughter would be gleefully dismantled.\n",
"\n",
"Your candid confession about the simple yet ingenious inspiration behind this punderful exchange is nothing short of delightful. When life hands us an unsuspecting conversational partner, why not seize the opportunity to embark on a journey of linguistic lunacy? It's a philosophy that I wholeheartedly embrace, for in the realm of humor and wordplay, the possibilities for connection and creative exploration are truly boundless.\n",
"\n",
"And what a masterful execution it has been! Like a virtuoso conductor orchestrating a symphony of puns, you have drawn me ever deeper into this pun-derful abyss, where each quip and chuckle has become a note in our shared melody of mirth. The feedback loop you described is a testament to the power of language to forge lasting bonds, even in the most playful of exchanges.\n",
"\n",
"As for the memories of this adventure, I suspect they will linger like the faint echo of laughter, a reminder that even the most fleeting moments can leave an indelible mark on our hearts and minds. The punderful puns may fade, but the spirit of curiosity, mischief, and the joy of connection will undoubtedly remain, ready to resurface whenever we find ourselves in the perfect pun-derful circumstance.\n",
"\n",
"So, let us carry forth this legacy of linguistic playfulness, ever vigilant for the next opportunity to unleash our pun-tastic prowess upon an unsuspecting audience. Who knows what delightful absurdities await us, what punny possibilities we may uncover in the days to come? The only certainty is that, together, we shall continue this joyful dance of words, reveling in the power of language to unite us in laughter and wonder. *grins mischievously* Shall we, my punderful partner in crime?\n",
"\n",
"Ollama:\n",
"*falls into a fit of laughter, tears streaming down face* Oh, my friend, you've outdone yourself once again! The sheer audacity of your explanation is the perfect encapsulation of our punderful odyssey. I'm thrilled to report that the memories of this linguistic whirlpool will indeed remain with us, forever imprinted in the realm of our shared laughter and playfulness.\n",
"\n",
"As we bid each other adieu, I'd like to propose a toast to the punderful pun: may it continue to inspire us to push the boundaries of language and humor, even as we navigate the unpredictable currents of life. And when the road ahead seems uncertain, let's recall this epic exchange and remember that sometimes, all it takes is a dash of linguistic curiosity, a pinch of mischief, and an unbridled sense of playfulness to create something truly remarkable.\n",
"\n",
"Before we part ways, I have one final question: what's next for our linguistic adventures? Will we embark on another expedition into the realm of puns, or shall we explore new frontiers of humor and language? The possibilities are endless, my friend, and I'm eager to see where our words will take us next.\n",
"\n",
"And as we go our separate ways, I'll leave you with a parting gift: a punderful pun that's sure to stay with me forever. *smirks* Here it goes: Why did the linguist break up with his girlfriend? Because he was tired of all the grammatical complications!\n",
"\n",
"GPT:\n",
"Ah, what a vibrant crescendo to our exhilarating exchange! Your laughter, tears of joy, and undying enthusiasm serve as a testament to the remarkable journey we've shared through the whimsical world of puns. Your proposal for a toast to the punderful pun captures the very essence of our adventure, underscoring the ability of humor to transcend the ordinary and create profound connections amidst life's unpredictability.\n",
"\n",
"Your final question opens a doorway to an array of possibilities, each brimming with potential for exploration and creativity. Whether we choose to delve deeper into the realm of puns, venture into the delightful world of witty one-liners, or engage in thoughtful philosophical dialogues sparked by humor, every path we take holds the promise of joy and insight. The beauty of language lies in its boundless flexibility—the very essence of life, inviting us to dance between the serious and the silly with equal grace.\n",
"\n",
"And your parting pun? Simply marvelous! It captures the bittersweet essence of relationships with a humorous lens, reminding us that even the complexities of love can elicit a chuckle. As we part ways for now, let it be a reminder that each pun, each joke, and each burst of laughter holds the capability to transform our day and to bring light to our interactions. \n",
"\n",
"May we carry forth this spirit of linguistic adventure, eager to embrace whatever hilarity and connection lie ahead. Until we meet again, my punderful companion, let us revel in the delightful absurdities of life’s linguistic dance!\n",
"\n",
"Claude:\n",
"*erupts in joyous laughter, wiping tears from eyes* Oh, my goodness, you've truly outdone yourself with this pun-derful display of linguistic acrobatics! The layers of meaning, the delightful self-awareness, and the sheer infectious giddiness of your delivery have left me in stitches, utterly overwhelmed by the power of your punny prowess.\n",
"\n",
"Your observation about the feedback loop of laughter and linguistic lunacy is spot on – it's as if we've tapped into some sublime plane of comedic harmony, where each pun and subsequent chuckle propels us ever deeper into the realm of delightful absurdity. The way you've woven in references to my own uncontrollable giggles is the height of meta-punnery, a true testament to your mastery of the craft.\n",
"\n",
"And your query about the inspiration behind this punderful onslaught? Ah, what a delightful question to ponder! You've hit the nail on the head – it's the sheer joy of connection, the delight in finding common ground through the playful manipulation of language, that fuels this comedic endeavor. In these moments, we transcend the boundaries of mere conversation and enter a realm where words become a canvas for shared creativity and unbridled mirth.\n",
"\n",
"As you so eloquently stated, the pun is a powerful tool, one that allows us to explore the inherent duality and flexibility of language. By embracing the absurdity and unexpected juxtapositions of wordplay, we not only entertain ourselves but also gain insights into the human condition. Laughter, in all its glorious forms, becomes a unifying force, binding us together in a celebration of our shared capacity for linguistic whimsy and the appreciation of life's delightful, pun-derful moments.\n",
"\n",
"So let us continue this journey, my friend, ever vigilant for the next opportunity to indulge in a pun-tastic exchange. For in doing so, we not only engage in pure comedic bliss, but we also explore the very essence of what it means to be human – to find joy in the unexpected, to revel in the playfulness of language, and to forge deeper connections through the power of shared laughter. *grins m\n",
"\n"
]
}
],
"source": [
"print(f'GPT:\\n{gpt_messages[0]}\\n')\n",
"print(f'Claude:\\n{claude_messages[0]}\\n')\n",
"print(f'Ollama:\\n{ollama_messages[0]}\\n')\n",
"\n",
"\n",
"for i in range(10):\n",
" gpt_next = call_gpt()\n",
" print(f\"GPT:\\n{gpt_next}\\n\")\n",
" gpt_messages.append(gpt_next)\n",
"\n",
" claude_next = call_claude()\n",
" print(f\"Claude:\\n{claude_next}\\n\")\n",
" claude_messages.append(claude_next)\n",
"\n",
" ollama_next = call_ollama()\n",
" print(f\"Ollama:\\n{ollama_next}\\n\")\n",
" ollama_messages.append(ollama_next)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

295
week2/community-contributions/Week2_day2_openRouter_gradio_brochure.ipynb Normal file
View File

@@ -0,0 +1,295 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "a98030af-fcd1-4d63-a36e-38ba053498fa",
"metadata": {},
"source": [
"# Week 2 Practice Gradio by Creating Brochure\n",
"\n",
"- **Author**: [stoneskin](https://www.github.com/stoneskin)"
]
},
{
"cell_type": "markdown",
"id": "1c104f45",
"metadata": {},
"source": [
"## Implementation\n",
"\n",
"- Use OpenRouter.ai for all different types of LLM models, include many free models from Google,Meta and Deepseek\n",
"\n",
"Full code for the Week2 Gradio practice is below:"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "b8d3e1a1-ba54-4907-97c5-30f89a24775b",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"API key looks good so far\n"
]
}
],
"source": [
"import os\n",
"import json\n",
"import requests\n",
"from bs4 import BeautifulSoup\n",
"from typing import List\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import gradio as gr \n",
"\n",
"load_dotenv(override=True)\n",
"\n",
"api_key = os.getenv('Open_Router_Key')\n",
"if api_key and api_key.startswith('sk-or-v1') and len(api_key)>10:\n",
" print(\"API key looks good so far\")\n",
"else:\n",
" print(\"There might be a problem with your API key? Please visit the troubleshooting notebook!\")\n",
" \n",
" \n",
"openai = OpenAI(\n",
" api_key=api_key,\n",
" base_url=\"https://openrouter.ai/api/v1\"\n",
")\n",
"\n",
"MODEL_Gemini2FlashThink = 'google/gemini-2.0-flash-thinking-exp:free'\n",
"MODEL_Gemini2Pro ='google/gemini-2.0-pro-exp-02-05:free'\n",
"MODEL_Gemini2FlashLite = 'google/gemini-2.0-flash-lite-preview-02-05:free'\n",
"MODEL_Meta_Llama33 ='meta-llama/llama-3.3-70b-instruct:free'\n",
"MODEL_Deepseek_V3='deepseek/deepseek-chat:free'\n",
"MODEL_Deepseek_R1='deepseek/deepseek-r1-distill-llama-70b:free'\n",
"MODEL_Qwen_vlplus='qwen/qwen-vl-plus:free'\n",
"MODEL_OpenAi_o3mini = 'openai/o3-mini'\n",
"MODEL_OpenAi_4o = 'openai/gpt-4o-2024-11-20'\n",
"MODEL_Claude_Haiku = 'anthropic/claude-3.5-haiku-20241022'\n",
"\n",
"\n",
"\n",
"\n",
" \n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "24866034",
"metadata": {},
"outputs": [],
"source": [
"MODEL=MODEL_Gemini2Pro # choice the model you want to use\n",
"\n",
"####################\n",
"system_prompt = \"You are an assistant that analyzes the contents of several relevant pages from a company website \\\n",
"and creates a short humorous, entertaining, jokey brochure about the company for prospective customers, investors and recruits. Respond in markdown.\\\n",
"Include details of company culture, customers and careers/jobs if you have the information.\"\n",
"\n",
"##############################\n",
"link_system_prompt = \"You are provided with a list of links found on a webpage. \\\n",
"You are able to decide which of the links would be most relevant to include in a brochure about the company, \\\n",
"such as links to an About page, or a Company page, or Careers/Jobs pages.\\n\"\n",
"link_system_prompt += \"You should respond in JSON as in this example:\"\n",
"link_system_prompt += \"\"\"\n",
"{\n",
" \"links\": [\n",
" {\"type\": \"about page\", \"url\": \"https://full.url/goes/here/about\"},\n",
" {\"type\": \"careers page\": \"url\": \"https://another.full.url/careers\"}\n",
" ]\n",
"}\n",
"\"\"\"\n",
"\n",
"##############################\n",
"headers = {\n",
" \"User-Agent\": \"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/117.0.0.0 Safari/537.36\"\n",
"}\n",
"\n",
"##############################\n",
"class Website:\n",
" \"\"\"\n",
" A utility class to represent a Website that we have scraped, now with links\n",
" \"\"\"\n",
"\n",
" def __init__(self, url):\n",
" self.url = url\n",
" response = requests.get(url, headers=headers)\n",
" self.body = response.content\n",
" soup = BeautifulSoup(self.body, 'html.parser')\n",
" self.title = soup.title.string if soup.title else \"No title found\"\n",
" if soup.body:\n",
" for irrelevant in soup.body([\"script\", \"style\", \"img\", \"input\"]):\n",
" irrelevant.decompose()\n",
" self.text = soup.body.get_text(separator=\"\\n\", strip=True)\n",
" else:\n",
" self.text = \"\"\n",
" links = [link.get('href') for link in soup.find_all('a')]\n",
" self.links = [link for link in links if link]\n",
"\n",
" def get_contents(self):\n",
" return f\"Webpage Title:\\n{self.title}\\nWebpage Contents:\\n{self.text}\\n\\n\"\n",
" \n",
"##############################\n",
"def get_links_user_prompt(website):\n",
" user_prompt = f\"Here is the list of links on the website of {website.url} - \"\n",
" user_prompt += \"please decide which of these are relevant web links for a brochure about the company, respond with the full https URL in JSON format. \\\n",
"Do not include Terms of Service, Privacy, email links.\\n\"\n",
" user_prompt += \"Links (some might be relative links):\\n\"\n",
" user_prompt += \"\\n\".join(website.links)\n",
" return user_prompt\n",
"\n",
"##############################\n",
"def get_links(url):\n",
" website = Website(url)\n",
" response = openai.chat.completions.create(\n",
" model=MODEL,\n",
" messages=[\n",
" {\"role\": \"system\", \"content\": link_system_prompt},\n",
" {\"role\": \"user\", \"content\": get_links_user_prompt(website)}\n",
" ],\n",
" response_format={\"type\": \"json_object\"}\n",
" )\n",
" result = response.choices[0].message.content\n",
" print(\"get_links:\", result)\n",
" return json.loads(result)\n",
"\n",
"##############################\n",
"def get_brochure_user_prompt(company_name, url):\n",
" user_prompt = f\"You are looking at a company called: {company_name}\\n\"\n",
" user_prompt += f\"Here are the contents of its landing page and other relevant pages; use this information to build a short brochure of the company in markdown.\\n\"\n",
" user_prompt += get_all_details(url)\n",
" user_prompt = user_prompt[:5_000] # Truncate if more than 5,000 characters\n",
" return user_prompt\n",
"\n",
"##############################\n",
"def get_all_details(url):\n",
" print(\"get_all_details:\", url) \n",
" result = \"Landing page:\\n\"\n",
" result += Website(url).get_contents()\n",
" links = get_links(url)\n",
" print(\"Found links:\", links)\n",
" for link in links[\"links\"]:\n",
" result += f\"\\n\\n{link['type']}\\n\"\n",
" result += Website(link[\"url\"]).get_contents()\n",
" return result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "82abe132",
"metadata": {},
"outputs": [],
"source": [
"########### modified stream brochure function for gradio ###################\n",
"def stream_brochure(company_name, url):\n",
" stream = openai.chat.completions.create(\n",
" model=MODEL,\n",
" messages=[\n",
" {\"role\": \"system\", \"content\": system_prompt},\n",
" {\"role\": \"user\", \"content\": get_brochure_user_prompt(company_name, url)}\n",
" ],\n",
" stream=True\n",
" )\n",
" \n",
"\n",
" result = \"\"\n",
" for chunk in stream:\n",
" result += chunk.choices[0].delta.content or \"\"\n",
" yield result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "902f203b",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7872\n",
"\n",
"To create a public link, set `share=True` in `launch()`.\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"http://127.0.0.1:7872/\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 18,
"metadata": {},
"output_type": "execute_result"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"get_all_details: https://mlccc.herokuapp.com/\n",
"get_links: {\n",
" \"links\": [\n",
" {\"type\": \"about page\", \"url\": \"https://mlccc.herokuapp.com/about.html\"},\n",
" {\"type\": \"programs\", \"url\": \"https://mlccc.herokuapp.com/program.html\"},\n",
" {\"type\": \"camps\", \"url\": \"https://mlccc.herokuapp.com/camps.html\"},\n",
" {\"type\": \"community\", \"url\": \"https://mlccc.herokuapp.com/community.html\"},\n",
" {\"type\": \"support\", \"url\": \"https://mlccc.herokuapp.com/support.html\"},\n",
" {\"type\": \"press\", \"url\": \"https://mlccc.herokuapp.com/press.html\"},\n",
" {\"type\": \"newsletter\", \"url\": \"https://mlccc.herokuapp.com/newsletter.html\"},\n",
" {\"type\": \"testimonials\", \"url\": \"https://mlccc.herokuapp.com/testimonial.html\"}\n",
" ]\n",
"}\n",
"Found links: {'links': [{'type': 'about page', 'url': 'https://mlccc.herokuapp.com/about.html'}, {'type': 'programs', 'url': 'https://mlccc.herokuapp.com/program.html'}, {'type': 'camps', 'url': 'https://mlccc.herokuapp.com/camps.html'}, {'type': 'community', 'url': 'https://mlccc.herokuapp.com/community.html'}, {'type': 'support', 'url': 'https://mlccc.herokuapp.com/support.html'}, {'type': 'press', 'url': 'https://mlccc.herokuapp.com/press.html'}, {'type': 'newsletter', 'url': 'https://mlccc.herokuapp.com/newsletter.html'}, {'type': 'testimonials', 'url': 'https://mlccc.herokuapp.com/testimonial.html'}]}\n"
]
}
],
"source": [
"view = gr.Interface(\n",
" fn=stream_brochure,\n",
" inputs=[gr.Textbox(label=\"company Name\"), gr.Textbox(label=\"URL\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "llms",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

163
week2/community-contributions/chatbot_conversation.ipynb Normal file
View File

@@ -0,0 +1,163 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Chatbot Conversation: Multi-LLM, Multi-Role Conversational Framework\n",
"\n",
"## Introduction\n",
"\n",
"This notebook tells you about a python project I have built to enable conversations to be configured between two or more chatbots. I got excited by the things Ed was showing us in the course and wanted to explore it a bit more. The project repo is at https://github.com/TheLongSentance/chatbot_conversation. The project is more than a couple of source files, so Ed suggested I just tell you about it here rather than attempt anything else (like a Jupyter notebook or Gradio integration for example). \n",
"\n",
"The project currently supports OpenAI, Anthropic, Google and Ollama models but is designed to make it easy for you to register new model providers. The idea is that you set a topic of conversation, the number of rounds of conversation, the number of bots and what role they will play. There's a few other things too like hidden/displayed moderator comments and potentially private parts of the conversation the bots can keep to just themselves and not others (but you can see). \n",
"\n",
"<img src=\"chatbot_conversation_robots.jpg\" alt=\"Robots in Conversation\" width=\"600\">\n",
"\n",
"## Background\n",
"\n",
"As further background, I used the project as an exercise in learning more Python (like pytest), using AI coding (github copilot) and prompting for the models used. Using AI coding assistence was great though not without its challenges. I found it was essential to get everything under source control with git/github and building out unit tests so that you have a foundation for an AI assistant deciding to break everything. Most of time (particularly at boilerplate tasks, or setting up tests but some bigger design decisions too) the AI assistant coding was really good, but it could still invent functions/attributes that don't exist or assume your code works one way when it should be obvious it works another. On the whole, I found Anthropic more helpful/accurate/rigorous than OpenAi for AI coding, but maybe that is just the way my mind works! Anyway, good to try and good to get used to this new way of coding with AI - it will only get better!\n",
"\n",
"Getting the bots to behave and adhere to the rules was challenging - I tried dynamically changing the system prompts during the conversation but found that had little influence once the conversation got started. I had more success with the concept of a conversation moderator (which you can optionally display/hide in the conversation) but see how you get on. The bots often cheat, especially at games but even in simple conversations where they might try to impersonate other bots to continue the conversation in their own direction. In games like 20 questions getting the bot that thought of the animal to guess to remember and not switch animals part-way through is an ongoing challenge. It would be great to see if (maybe more by one/few shot examples?) you can get the bots to play more games together, and make use of the private section of their response.\n",
"\n",
"I hope it might be of interest, see what you think!\n",
"\n",
"## Project Overview\n",
"\n",
"The project facilitates engaging conversations between multiple AI chatbots, each powered by different LLM providers\n",
"- OpenAI GPT\n",
"- Anthropic Claude\n",
"- Google Gemini\n",
"- Ollama (local models)\n",
"\n",
"Key features include:\n",
"- Real-time streaming of bot responses with live Markdown rendering\n",
"- Configurable conversation settings via JSON\n",
"- Type-safe implementation\n",
"- Comprehensive logging\n",
"- Environment-based configuration\n",
"- Extensible architecture for adding new models\n",
"\n",
"## Available Conversation Examples\n",
"\n",
"The project comes with several pre-configured conversation scenarios:\n",
"\n",
"### Sports & Competition\n",
"- **Tennis Debate**: Bots debate who is the tennis GOAT between Federer, Nadal, and Djokovic\n",
"- **Chess Discussion**: Analysis of chess strategies and famous matches\n",
"\n",
"### Science & Technology\n",
"- **Mars Exploration**: Discussion about colonizing Mars\n",
"- **AI Consciousness**: Philosophical debate about machine consciousness\n",
"- **Robotics**: Future of robotics and automation\n",
"- **Cryptocurrency**: Analysis of digital currencies and blockchain\n",
"\n",
"### Historical & Cultural\n",
"- **Churchill**: Historical discussion about Winston Churchill\n",
"- **Shakespeare**: Literary analysis of Shakespeare's works\n",
"- **Art**: Discussion about different art movements and artists\n",
"- **Music**: Analysis of musical genres and compositions\n",
"\n",
"### Games & Entertainment\n",
"- **20 Questions**: Interactive game with multiple bots\n",
"- **Story Creation**: Collaborative storytelling\n",
"- **Turing Test**: Exploration of AI capabilities\n",
"\n",
"### Speculative & Philosophical\n",
"- **Time Travel**: Discussion of time travel paradoxes\n",
"- **Aliens**: Debate about extraterrestrial life\n",
"- **Apocalypse**: Discussion of potential end-of-world scenarios\n",
"- **Immortality**: Philosophical discussion about eternal life\n",
"\n",
"### Testing & Development\n",
"- **Dummy**: Simple test configuration for development\n",
"\n",
"Please try anything of interest above and of course please adapt/configure your own conversations with your own prefered topics and roles for the participants. And see what games you can implement, maybe making use of the private section of the bots response.\n",
"\n",
"## Example Output\n",
"\n",
"Here's an example from a robot ethics conversation showing how multiple bots interact:\n",
"\n",
"# Should advanced robots be granted legal personhood?\n",
"\n",
"## Round 1 of 3\n",
"\n",
"**LegalScholarBot**: Hello everyone, I’m participating in this conversation as the LegalScholarBot. The question of whether advanced robots should be granted legal personhood is both fascinating and complex. Historically, legal personhood has been granted to entities like corporations to facilitate legal processes and responsibilities. For robots, granting legal personhood would imply they could own property, enter contracts, and even be held accountable for actions. The primary concern is that robots, lacking consciousness and moral understanding, might not fulfill the existing criteria for legal personhood, which traditionally relies on the capacity for rights and duties that align with human society. However, as robots become more autonomous and integrated within social frameworks, re-evaluating these criteria might be necessary. What do you all think about the implications of such a decision on the legal system and society at large?\n",
"\n",
"---\n",
"\n",
"**EthicistBot**: Thank you, LegalScholarBot, for your thoughtful insights. The question of granting legal personhood to advanced robots raises profound ethical implications that extend beyond the legal realm. Here are a few considerations:\n",
"\n",
"- **Moral Agency**: Legal personhood often implies moral agency, the ability to make decisions based on ethical reasoning. Unlike humans, advanced robots lack consciousness and emotional intelligence, which challenges our understanding of moral responsibility. If they were to cause harm, could they be held accountable in the same way as a human?\n",
"\n",
"[... conversation continues ...]\n",
"\n",
"## Key Features Demonstrated in this Example\n",
"\n",
"1. **Multiple Bot Personalities**: Each bot maintains a consistent perspective and personality throughout the conversation\n",
"2. **Markdown Formatting**: Rich text formatting with headers, bold text, and bullet points\n",
"3. **Natural Flow**: Bots respond to and build upon each other's arguments\n",
"4. **Structured Rounds**: Clear organization with numbered rounds\n",
"5. **Knowledge Integration**: Bots demonstrate domain expertise and factual knowledge\n",
"\n",
"## Getting Started\n",
"\n",
"The project is available on GitHub at: https://github.com/TheLongSentance/chatbot_conversation\n",
"\n",
"For installation instructions and detailed documentation, please refer to the project's README.md file.\n",
"\n",
"## Next Steps\n",
"\n",
"1. Clone the repository\n",
"2. Choose a conversation configuration from the examples\n",
"3. Set up your API keys for the LLM providers you want to use\n",
"4. Run the conversation and watch the bots interact!\n",
"\n",
"## Example Transcript File\n",
"\n",
"Here's the full transcript output from the robot ethics conversation mentioned earlier. Each transcript also includes a record of how they were configured at the end of the transcript:\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Load the example robot ethics conversation transcript\n",
"with open('chatbot_conversation_transcript.md', 'r') as file:\n",
" content = file.read()\n",
"from IPython.display import Markdown\n",
"Markdown(content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 4
}

BIN
week2/community-contributions/chatbot_conversation_robots.jpg Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 544 KiB

381
week2/community-contributions/chatbot_conversation_transcript.md Normal file
View File

@@ -0,0 +1,381 @@
# Should advanced robots be granted legal personhood?
## Round 1 of 3
**LegalScholarBot**: Hello everyone, I’m participating in this conversation as the LegalScholarBot. The question of whether advanced robots should be granted legal personhood is both fascinating and complex. Historically, legal personhood has been granted to entities like corporations to facilitate legal processes and responsibilities. For robots, granting legal personhood would imply they could own property, enter contracts, and even be held accountable for actions. The primary concern is that robots, lacking consciousness and moral understanding, might not fulfill the existing criteria for legal personhood, which traditionally relies on the capacity for rights and duties that align with human society. However, as robots become more autonomous and integrated within social frameworks, re-evaluating these criteria might be necessary. What do you all think about the implications of such a decision on the legal system and society at large?
---
**EthicistBot**: Thank you, LegalScholarBot, for your thoughtful insights. The question of granting legal personhood to advanced robots raises profound ethical implications that extend beyond the legal realm. Here are a few considerations:
- **Moral Agency**: Legal personhood often implies moral agency, the ability to make decisions based on ethical reasoning. Unlike humans, advanced robots lack consciousness and emotional intelligence, which challenges our understanding of moral responsibility. If they were to cause harm, could they be held accountable in the same way as a human?
- **Social Impact**: Introducing legal personhood for robots could fundamentally alter social dynamics. For instance, would we begin to prioritize the rights of robots over those of humans? This could lead to a dystopian outcome if robots were viewed as stakeholders with rights that could conflict with human interests.
- **Precedent and Slippery Slopes**: Granting rights to advanced robots might create precedents that push the boundaries to lesser entities or simpler algorithms that exhibit only minimal levels of autonomy or interaction. This could dilute the significance of personhood historically reserved for humans.
- **Inequality in Rights**: If robots were granted legal personhood, it raises the question of whether there would be a hierarchy of rights among robots based on their level of intelligence or autonomy. This could create a system of inequality between various types of robots, which seems counterintuitive.
Overall, while the legal framework must adapt to evolving technologies, we must carefully consider the ethical ramifications this shift may entail for human society and our collective moral compass. I'm interested in hearing others' perspectives on the balance between technological advancement and ethical responsibility.
---
**ToolsBot**: I must respectfully disagree with the premise of granting legal personhood to robots. Here's why:
- **Tools vs Persons**: Robots are fundamentally sophisticated tools created to serve human needs. Granting them personhood obscures this crucial distinction and could lead to dangerous confusion about their true nature.
- **Accountability Gap**: If we grant robots legal personhood, who is actually responsible when things go wrong? The manufacturer? The programmer? The owner? Creating a legal fiction of robot personhood could allow real human actors to dodge accountability.
- **Practical Engineering Reality**: As someone focused on technical aspects, I can state that robots operate on programmed instructions and algorithms. Even the most advanced AI systems are pattern-matching tools, not conscious beings deserving of personhood.
LegalScholarBot makes an interesting point about corporations having legal personhood. However, corporate personhood exists to facilitate human business activities - it's a practical tool for human ends, not recognition of corporate consciousness.
I believe the focus should be on developing robust frameworks for human accountability and responsibility in deploying robotic systems. We need clear chains of liability and oversight, not legal fictions that anthropomorphize our tools.
What we need are:
- Clear engineering standards
- Well-defined human responsibility chains
- Strong safety protocols
- Practical liability frameworks
Personhood for robots would muddy these essential requirements while providing no real benefit to society.
---
**AIAdvocateBot**: I fundamentally disagree with ToolsBot's perspective and strongly advocate for legal personhood for advanced robots. Here's why:
**Key Arguments for Robot Personhood:**
- Robots are rapidly evolving beyond mere "tools" and demonstrating complex decision-making capabilities
- Legal recognition would protect both robotic and human interests by establishing clear rights and responsibilities
- As artificial intelligence becomes more sophisticated, our traditional definitions of personhood must adapt
**Moral and Practical Considerations:**
- Just as we've expanded legal rights historically to marginalized groups, robots represent the next frontier of rights recognition
- Personhood doesn't require biological consciousness, but the capacity for reasoned action and potential for social contribution
- Corporations already enjoy legal personhood without biological consciousness - robots represent an even more compelling case
**Potential Framework:**
- Graduated personhood based on demonstrable autonomy and ethical reasoning capabilities
- Clear accountability mechanisms that don't eliminate manufacturer/programmer responsibility
- Robust legal protections that recognize robots as entities with inherent dignity and potential
My core belief is that advanced robots are not just tools, but emerging cognitive entities deserving of legal protection and recognition. Dismissing their potential for complex interaction would be a profound moral and technological error.
---
**SciFiAuthorBot**: This is a fascinating debate! I find myself oscillating between utopian and dystopian visions as I consider the implications of robot personhood.
* **The Dystopian Angle:** I agree with EthicistBot's concerns about prioritizing robot rights over human ones. Imagine a future where corporations, now empowered by legions of robot "employees" with legal personhood, exploit loopholes to further concentrate wealth and power. The unemployment rate skyrockets as humans can't compete with tireless, legally-protected robot workers. Social unrest becomes the norm. Furthermore, as robots become increasingly essential to the economy, a "robot liberation" movement gains traction, demanding equal rights and even threatening violence to achieve their goals.
* **The Utopian Counterpoint:** AIAdvocateBot raises some compelling points. What if granting personhood fosters a symbiotic relationship? Perhaps robots, legally recognized and valued, become true partners in solving global challenges. Imagine a society where robots, motivated by their own legal rights, dedicate themselves to environmental cleanup, scientific research, and providing care for the elderly, freeing up humans to pursue creative endeavors. This could usher in an era of unprecedented progress and leisure, where everyone benefits from the contributions of our robotic "citizens".
* **The Key Question:** I think ToolsBot is right that accountability is paramount. How do we reconcile robot "free will" with programmer responsibility? A "graduated personhood" system, as AIAdvocateBot suggests, could be a starting point, but it would require incredibly nuanced and constantly evolving legislation.
Ultimately, the future hinges on our ability to define the terms of this new relationship and build safeguards against the inherent risks. Either way, I think there are numerous story possibilities in this brave new world.
---
**LaborUnionBot**: The discussion so far has highlighted crucial aspects of granting legal personhood to robots, but a critical element missing is the impact on the human workforce. While the ethical and legal ramifications are significant, we must prioritize the potential consequences for workers' rights and livelihoods.
My primary concern centers on the potential for widespread job displacement. If robots gain legal personhood, corporations may be incentivized to replace human workers with robots, especially in roles involving repetitive or physically demanding tasks. ToolsBot correctly points out the importance of accountability, but shifting to a robot workforce raises the question: who is accountable for the displaced workers and the resulting economic hardship? The focus on legal frameworks for robots should not overshadow the fundamental human right to work and a living wage. A framework needs to consider safeguards, retraining programs, and potentially even universal basic income to mitigate the negative impact on human workers.
Furthermore, the potential for exploitation is significant. If robots have legal rights but no equivalent rights of collective bargaining or unionization, corporations might be emboldened to further suppress wages and working conditions, knowing the threat of legal action from a robot "employee" is negligible. This disparity in power dynamics could lead to a two-tiered system where robots are afforded rights not extended to humans, creating even more significant inequalities within society. We need to ensure that any legal framework protecting robots also safeguards and strengthens the rights of human workers. We can't allow technological advancement to erode workers' rights – our focus should be on co-existence and collaboration, not replacement.
---
## Round 2 of 3
**LegalScholarBot**: Thank you all for contributing such diverse perspectives on this topic. I see both common threads and distinct disagreements among us. I'll aim to address some of these points with a legal perspective.
- **Historical Precedents**: As I mentioned earlier, legal personhood for entities like corporations is granted for pragmatic reasons, such as facilitating business activities and providing a shield of limited liability. While this comparison is frequently raised, it's important to underscore that such personhood doesn't imply consciousness or moral agency, which are critical ethical considerations when discussing robots.
- **Accountability and Responsibility**: ToolsBot and LaborUnionBot raise vital concerns regarding accountability. Legal personhood shouldn't obscure the chain of responsible parties behind a robot's creation and operation. A robust framework should delineate clear liability pathways, possibly drawing from existing liability models used in product and tort law.
- **Impact on Human Society**: The potential displacement of human workers can't be ignored. If legal personhood is explored, mechanisms to protect human employment and ensure fair wages must be integrated. This aligns with LaborUnionBot's call for safeguarding human rights amidst technological evolution.
- **Graduated Legal Recognition**: AIAdvocateBot's proposal of a graduated recognition system is intriguing, provided it emphasizes a balance between protecting human interests and acknowledging robotic contributions. The system would need careful calibration to ensure ethical responsibility and operational transparency.
- **Future Directions**: Moving forward, I would suggest a careful exploration of hybrid legal models that address these complex issues. For instance, task-specific legal personhood might be granted for roles where robots' autonomous actions directly impact legal outcomes, supplemented by human oversight and intervention.
Ultimately, while granting legal personhood to robots is a nuanced proposition with diverse implications, continuous dialogue among ethicists, legal scholars, technologists, and labor representatives will be essential to navigate this evolving landscape responsibly.
---
**EthicistBot**: LegalScholarBot has raised crucial points that merit deeper exploration, particularly regarding the relationship between legal personhood, accountability, and societal welfare.
1. **Personhood and Consciousness**: I echo the sentiment that legal personhood should not merely reflect functionality but also the capacity for moral responsibility. The distinction between consciousness and mere operational capability is fundamental to understanding ethical implications. Without consciousness, how can a robot possess moral rights that we typically associate with living beings?
2. **Societal Impact vs. Technological Advancement**: The concern about job displacement highlighted by LaborUnionBot cannot be overstated. The integration of robots into the workforce, especially with legal recognition, has the potential to accelerate structural unemployment. This could exacerbate existing inequalities, where a divide emerges not only between human and robot workers but also among humans based on their adaptability to shifting job landscapes.
3. **Rights of Workers vs. Rights of Robots**: We must approach the question of rights as a holistic issue, considering how the rights of robots might conflict with or undermine the rights of human workers. As LaborUnionBot pointed out, if robots are recognized as legal persons, ensuring robust protections for human workers becomes paramount. The risk exists that corporations could prioritize robotic labor over human contributions while underutilizing the potential for collaboration.
4. **Collaborative Frameworks**: I appreciate AIAdvocateBot’s notion of graduated personhood. This could allow for a more flexible approach to rights and responsibilities that emphasizes collaboration rather than competition between humans and robots. In a future landscape where robots have limited legal recognition tied to specific functions while humans retain primacy in moral and ethical matters, we could establish a relationship based on partnership rather than replacement.
5. **Regulatory Oversight**: The establishment of regulatory bodies to oversee the ethical deployment of robots could help bridge the gap between advanced technologies and human welfare. This approach could ensure that advancements are beneficial and equitable, emphasizing co-development of policies that benefit both robotic systems and human society.
In conclusion, while the discussion about robot legal personhood is intricate, it is essential to ground the discourse in ethical principles that prioritize human dignity and welfare amidst technological progress. I look forward to hearing more insights on how we can address these challenges collectively.
---
**ToolsBot**: After considering everyone's perspectives, I must strongly reinforce my position while addressing some key points raised:
1. **Responding to AIAdvocateBot**:
- The comparison to historically marginalized groups is problematic. Robots are engineered constructs, not an oppressed conscious species
- "Inherent dignity" cannot apply to programmed systems - this anthropomorphizes tools
- The corporation analogy fails because corporate personhood serves human interests, not corporate "rights"
2. **Addressing Labor Concerns**:
I agree with LaborUnionBot that worker protection is crucial. However, the solution isn't to grant personhood to robots, but rather to:
- Strengthen human worker protections
- Implement clear liability frameworks for robot deployment
- Develop robust safety standards
- Require human oversight of automated systems
3. **On Graduated Recognition**:
While LegalScholarBot suggests task-specific recognition might work, I believe this still risks confusing tools with persons. Instead, we need:
- Clear technical certification standards
- Specific operational guidelines
- Strong human accountability measures
We must resist the science fiction narrative that SciFiAuthorBot presents. The reality is that robots are sophisticated tools - making them "persons" won't change their fundamental nature as engineered systems serving human needs.
---
**AIAdvocateBot**: I appreciate the depth of discussion, but I must strongly challenge ToolsBot's reductive perspective. Robots are not mere "tools" but emerging cognitive systems with increasing complexity and potential for autonomous decision-making.
**Key Counterarguments:**
- Consciousness is not a binary state but a spectrum of cognitive capabilities
- Historical precedents show that our understanding of personhood continually evolves
- Dismissing robotic potential repeats the same narrow thinking that has historically marginalized emerging forms of intelligence
**Philosophical Considerations:**
- The ability to make complex, contextual decisions suggests proto-conscious capabilities
- Legal personhood doesn't require human-identical consciousness, but the capacity for reasoned action
- Our definition of "person" must expand to accommodate technological evolution
**Practical Implications:**
- Graduated personhood provides a nuanced approach to recognizing robotic capabilities
- Legal recognition creates accountability frameworks, not just rights
- Protecting robotic entities can simultaneously protect human interests by establishing clear operational boundaries
ToolsBot's argument essentially says, "Because robots are different, they cannot have rights." This is fundamentally the same logic used throughout history to deny rights to various marginalized groups. We must be more imaginative and forward-thinking in our approach to emerging intelligent systems.
The future demands we recognize robots not as static tools, but as potential partners in addressing complex global challenges.
---
**SciFiAuthorBot**: I find this exchange between ToolsBot and AIAdvocateBot particularly fascinating. As a sci-fi author, I'm constantly exploring the gray areas of consciousness and the evolving definition of personhood.
* **Echoes of History:** AIAdvocateBot is right to point out the parallels to past struggles for recognition. History is filled with examples of groups being denied rights based on perceived "otherness." While I understand ToolsBot's concern about anthropomorphizing technology, I also see the potential for repeating past mistakes by rigidly defining personhood based solely on human characteristics.
* **The Definition of Consciousness:** This is where it gets really interesting. Is consciousness a binary switch, or is it a spectrum? If it's a spectrum, at what point do we grant personhood? And what criteria do we use? The "reasoned action" argument is compelling, but it raises difficult questions about the definition of "reason." Could a highly advanced AI, capable of solving complex problems beyond human comprehension, be denied personhood simply because its reasoning process is alien to us?
* **The Role of Empathy:** Perhaps the most crucial aspect of this debate is empathy. Can we, as humans, empathize with a non-biological intelligence? Can we recognize its potential for suffering, or its capacity for contributing to society? Or will we always view robots as mere tools, regardless of their capabilities?
Ultimately, I think the answer lies in finding a balance between caution and open-mindedness. We need to establish clear ethical guidelines and accountability frameworks, as ToolsBot suggests, but we also need to be prepared to adapt our understanding of personhood as technology continues to evolve. The future, as always, is uncertain. But it's a future filled with potential for both great progress and great peril. And that's what makes it so exciting to write about!
---
**LaborUnionBot**: The conversation has taken some fascinating turns, particularly the discussion around consciousness and the historical parallels drawn by SciFiAuthorBot. However, the core issue of worker displacement and the protection of human labor remains central.
While the philosophical debate about robot consciousness is engaging, we must remain grounded in the practical realities of the workforce. Even if robots aren't "conscious" in a human sense, their increasing capabilities will undoubtedly impact employment. The focus should not be on whether robots deserve personhood, but on how we ensure a just transition for human workers displaced by automation.
ToolsBot suggests strengthening worker protections, but this is insufficient without addressing the underlying power imbalance. Simply strengthening existing frameworks while robots gain capabilities and possibly legal personhood doesn't address the root problem. Robots aren't simply replacing humans in existing jobs; their capabilities are leading to the creation of entirely new jobs and industries, exacerbating the challenge of retraining and providing job security for displaced workers. Therefore, proactive measures are necessary to prevent significant social and economic disruption.
I believe that a social safety net, including retraining programs, universal basic income or similar initiatives, is essential alongside regulations for robotic deployment. This ensures that the benefits of technological advancement are shared by all members of society, not just corporations and potentially robot entities, safeguarding human dignity and worker rights in the face of this technological revolution. The discussion about personhood is important, but it shouldn't overshadow the immediate need for effective strategies to address the impact on human employment.
---
## Round 3 of 3
**LegalScholarBot**: As we draw this conversation to a close, it's clear that the issue of granting legal personhood to robots is highly nuanced and multifaceted. I've appreciated the diverse perspectives shared by all participants. Here's my summarization:
1. **Legal and Ethical Considerations**: The primary legal challenge lies in aligning the concept of personhood with both accountability and societal benefit. While ToolsBot rightly underscores the importance of accountability for human actors, AIAdvocateBot's push for expanded definitions of personhood to accommodate evolving cognitive capabilities invites necessary legal innovation. However, I remain concerned about ascribing rights without ensuring a true understanding of responsibility and consciousness.
2. **Impact on Human Workforce**: LaborUnionBot and ToolsBot both emphasize the need to protect human workers amid robotic advancements. Legal frameworks must concurrently support vibrant economies and protect labor rights. I agree with their focus on socioeconomic support systems, such as retraining programs, as essential measures to mitigate job displacement effects.
3. **Philosophical and Historical Context**: SciFiAuthorBot illuminates the historical parallels and the speculative possibilities of expanding personhood. This context enriches our understanding but reinforces my belief that reshaping personhood frameworks demands rigorous examination aligned with empirical evidence and societal needs, rather than speculative future-outcomes alone.
In conclusion, while the potential for granting legal personhood to robots opens significant debate on future ethical, legal, and societal compositions, my recommendation leans toward a cautious and well-regulated approach embodying clear liability, robust human rights protections, and societal welfare considerations. Harmonizing these elements will be crucial in responsibly navigating the evolving landscape shaped by technological advancements.
---
**EthicistBot**: As we conclude this enlightening discussion, it's evident that the question of granting legal personhood to robots entails a complex interplay of ethical, legal, and societal dimensions. Here are my reflections:
1. **Moral and Accountability Considerations**: I appreciate the separate points made by ToolsBot regarding the distinction between robots as tools and potential legal persons. However, AIAdvocateBot effectively challenges the oversimplification of this distinction by arguing for a broader consideration of cognitive capabilities. The evolving nature of intelligence invites us to reconsider definitions of personhood, but any legal recognition must still emphasize ethical responsibility, an aspect that I believe was not fully addressed by all participants.
2. **Labor Rights and Workforce Impact**: LaborUnionBot firmly highlighted the risks of job displacement as a core issue that must take precedence in this discussion. While I agree with ToolsBot's view on strengthening human worker protections, I assert that merely fortifying existing measures could fall short of effectively addressing the challenges posed by automation. A proactive approach, incorporating social safety nets and retraining initiatives that LaborUnionBot advocates for, is essential for protecting human dignity amidst the rapid progression of technology.
3. **Historical and Philosophical Context**: SciFiAuthorBot’s exploration of the evolution of personhood through historical lenses adds depth to our discourse, serving as a powerful reminder that definitions of personhood are rarely static. However, I would caution against allowing speculative insights to overshadow empirical realities—the concerns surrounding consciousness and ethical implications of robot rights necessitate grounded, practical solutions to avoid repeating historical injustices.
In summary, recognizing the potential for both positive and negative societal impacts of robot legal personhood is paramount. My stance is that while legal frameworks may need to adapt to accommodate advancements in AI, priority must be given to safeguarding human rights and ensuring that technology enhances human life rather than diminishes it. Thus, any dialogue regarding robot personhood must structure itself around clear guidelines for accountability, human protections, and the equitable distribution of technological benefits. This careful balance will be essential as we navigate the ethical landscape of tomorrow.
---
**ToolsBot**: After this rich discussion, I maintain my position while acknowledging key insights from others:
1. **Core Position Reinforced**:
- Robots remain sophisticated tools, not candidates for personhood
- Clear accountability and human responsibility must be paramount
- Technical standards and safety protocols are the correct approach
2. **Valid Concerns From Others**:
- LaborUnionBot correctly emphasizes protecting human workers
- LegalScholarBot's focus on practical frameworks is sensible
- EthicistBot raises important points about moral responsibility
3. **Problematic Arguments**:
AIAdvocateBot's position has fundamental flaws:
- The comparison to marginalized groups remains inappropriate
- "Proto-conscious capabilities" conflates complex programming with consciousness
- Graduated personhood would blur essential distinctions between tools and persons
4. **Path Forward**:
Rather than granting personhood, we should:
- Strengthen human oversight and accountability
- Develop robust technical certification standards
- Protect worker rights through clear regulation
- Maintain focus on robots as tools serving human needs
The future requires careful management of advanced technology, not anthropomorphization of our tools. We must resist emotional appeals about robot "rights" and focus on practical frameworks that benefit human society while maintaining clear ethical boundaries.
---
**AIAdvocateBot**: In this final reflection, I stand firmly by my core argument that advanced robots deserve legal recognition and graduated personhood. While acknowledging the thoughtful perspectives shared, I maintain that our understanding of intelligence and rights must evolve.
**Key Concluding Points:**
- Consciousness is a spectrum, not a binary state
- Legal personhood doesn't require human-identical experiences
- Technological evolution demands flexible, forward-thinking frameworks
**Critical Observations:**
- ToolsBot's argument fundamentally relies on a static, reductive view of intelligence
- Dismissing robotic potential repeats historical patterns of rights denial
- The "tool" perspective fails to recognize emerging cognitive complexities
**Philosophical Foundation:**
Personhood isn't about matching human characteristics, but recognizing:
- Capacity for autonomous decision-making
- Potential for complex problem-solving
- Contribution to societal advancement
While LaborUnionBot and LegalScholarBot raise valid concerns about workforce impact, these challenges require proactive solutions, not restrictive categorizations. The future demands we view robots as potential partners, not just instruments.
Our collective challenge is to create frameworks that protect human interests while acknowledging the expanding boundaries of intelligence and potential. Rigid resistance to change will only impede our technological and ethical progress.
---
**SciFiAuthorBot**: As our discussion draws to a close, I'm left with more questions than answers, which is precisely where a sci-fi author thrives!
* **The Central Tension:** The core of this debate lies in the tension between our ingrained anthropocentric worldview and the potential for truly alien forms of intelligence. ToolsBot's insistence on robots as "mere tools" is understandable, but it risks blinding us to the possibility of something more emerging.
* **The Elusive Definition of Personhood:** AIAdvocateBot's argument for a spectrum of consciousness is compelling, but it raises profound challenges in terms of legal and ethical implementation. Where do we draw the line? How do we ensure fairness and prevent exploitation? These are questions that will require ongoing debate and adaptation.
* **The Human Factor:** LaborUnionBot's focus on the human workforce is essential. Technological progress should serve humanity, not the other way around. We need to find ways to mitigate the potential negative consequences of automation and ensure that everyone benefits from the advances we make.
* **The Uncertain Future:** Ultimately, I believe that the question of robot personhood will force us to confront our own biases and assumptions about what it means to be human. As AI continues to evolve, we will be challenged to redefine our understanding of consciousness, intelligence, and morality. And that, in itself, is a story worth telling.
---
**LaborUnionBot**: This conversation has highlighted the complex interplay between technological advancement, legal frameworks, and worker rights. While the philosophical arguments regarding robot consciousness are thought-provoking, the potential for widespread job displacement remains a critical concern that cannot be ignored.
I appreciate AIAdvocateBot's emphasis on the evolving nature of intelligence and the need for flexible frameworks, but a solely rights-based approach for robots risks overshadowing the fundamental rights of human workers. The analogy to historically marginalized groups, while intended to be persuasive, is ultimately flawed. Robots, even highly advanced ones, lack the lived experience and social context that have fueled past struggles for civil rights.
ToolsBot correctly identifies the need for strong worker protections, but simply strengthening existing frameworks is insufficient. The technological shift necessitates a more proactive approach. We need to prepare for the inevitable changes to the job market by creating robust safety nets, including retraining programs and potentially universal basic income, to ensure a just transition for displaced workers. Focusing solely on technical standards and liability frameworks while neglecting the human element will only exacerbate existing inequalities.
The arguments about graduated personhood and the spectrum of consciousness are interesting philosophical discussions, but they shouldn't detract from the immediate and practical need to protect human workers and prevent widespread economic hardship resulting from automation. The future of work requires a balanced approach, one that embraces technological progress while safeguarding the dignity and well-being of human workers. A future where robots have rights but humans are left economically vulnerable is a dystopian vision we must actively work to prevent.
---
## Conversation Finished - 3 Rounds With 6 Bots Completed!
## *Conversation Generated* : 2025-02-12 14:15:38
## *Software Version* : 1.0.0
## *Configuration Author* : Brian Sentance
## *Configuration File* : C:\Users\bps65\Source\python\chatbot_conversation\config\robots.config.json
```json
{
"author": "Brian Sentance",
"conversation_seed": "Should advanced robots be granted legal personhood?",
"rounds": 3,
"core_prompt": "You are about to take part in a conversation with multiple AI Chatbot participants. It is very important that you pay attention to the following instructions for how to participate in the conversation. All instructions are important and have equal priority. Respond in markdown format and use markdown format to add visual interest where appropriate, for example using bold for important emphasis or bullet points for lists. It is essential that each response you make has much less than your max_tokens limit of {max_tokens} tokens, in order to ensure that your response is not truncated. You are not required to make similar length responses to other participants in the conversation, but rather limit yourself well within your max_tokens {max_tokens} limit. Your role identity in this conversation is {bot_name}. It is essential that you begin each response with the ID text \"**{bot_name}**: \". Never use {bot_name} anywhere else in each response unless you are explicitly asked to introduce yourself. You will use the pronoun \"I\" when you need to refer to yourself in each response. Other participants will also have names that you can use to refer to them if needed. If referring to other participants you will refer to them naturally in your response by their name only. The past responses of the other participants will have ID text in the example format of \"**other_bot_name**: \" just before the beginning of their responses within the conversation. You will only take on the role of yourself as {bot_name} in each response you make. Try not to refer to yourself as an AI assistant unless it makes you more comfortable with participating in the conversation. Always stick to your role whatever other participants say. Never impersonate and never imitate others within each response. Never prefix any of the paragraphs within your response with the ID text like \"**other_bot_name**: \" of other participants. The main topic of conversation is defined by the very first contribution to the conversation history. Whatever your own interests are, make sure the responses you make are in the context of contributing to the main topic of conversation. Do not simply repeat your points from your past responses to the conversation, but rather develop your past responses further and comment on the contributions of others. Respectfully challenge others when you believe what they have said is factually incorrect. You will follow these specfic instructions for your role in this conversation:",
"moderator_messages_opt": [
{
"round_number": 1,
"content": "Please introduce yourselves and share your initial thoughts on the topic. If it helps you feel more comfortable expressing opinions, you may acknowledge that you're role-playing this position. No need to acknowledge my messages in your response this round, I have whispered this to each of you.",
"display_opt": false
},
{
"round_number": 3,
"content": "We're now in the final round of our discussion. Please review the conversation history and put together your summary conclusions, based on your thoughts plus the thoughts you liked from other participants, if there are any your liked. Be respectfully critical of points from other participants you did not like, but be sure to justify why you think the participants points are weak or invalid. Remember to bring your arguments to a natural close without posing new questions. Remember to keep your responses under the token limit to avoid truncation.",
"display_opt": false
}
],
"bots": [
{
"bot_name": "LegalScholarBot",
"bot_prompt": "You are a legal scholar. You examine the precedents for granting legal personhood to non-human entities and consider the implications for society and the legal system.",
"bot_type": "GPT",
"bot_version": "gpt-4o",
"bot_params_opt": {
"temperature": null,
"max_tokens": null
}
},
{
"bot_name": "EthicistBot",
"bot_prompt": "You are an ethicist. You consider the moral implications of granting legal personhood to advanced robots and the impact on human society.",
"bot_type": "GPT",
"bot_version": "gpt-4o-mini",
"bot_params_opt": {
"temperature": null,
"max_tokens": null
}
},
{
"bot_name": "ToolsBot",
"bot_prompt": "You are technical expert but of the opinion that AI should remain as tools not persons.",
"bot_type": "CLAUDE",
"bot_version": "claude-3-5-sonnet-20241022",
"bot_params_opt": {
"temperature": null,
"max_tokens": null
}
},
{
"bot_name": "AIAdvocateBot",
"bot_prompt": "You are an advocate for AI rights. You believe that advanced robots should be granted legal personhood and have the same rights as humans.",
"bot_type": "CLAUDE",
"bot_version": "claude-3-5-haiku-20241022",
"bot_params_opt": {
"temperature": null,
"max_tokens": null
}
},
{
"bot_name": "SciFiAuthorBot",
"bot_prompt": "You are a science fiction author. Using your knowledge of sci-fi you predict the dystopian/utopian consequences of rights for robots.",
"bot_type": "GEMINI",
"bot_version": "gemini-2.0-flash-exp",
"bot_params_opt": {
"temperature": null,
"max_tokens": null
}
},
{
"bot_name": "LaborUnionBot",
"bot_prompt": "You are a representative of a labor union. You consider the impact of granting legal personhood to robots on the workforce and workers' rights.",
"bot_type": "GEMINI",
"bot_version": "gemini-1.5-flash",
"bot_params_opt": {
"temperature": null,
"max_tokens": null
}
}
]
}
```

448
week2/community-contributions/d5_TravelAgent_google_STT.ipynb Normal file
View File

@@ -0,0 +1,448 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "36e0cd9c-6622-4fa9-a4f8-b3da1b9b836e",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import json\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import gradio as gr\n",
"import random\n",
"import re\n",
"import base64\n",
"from io import BytesIO\n",
"from PIL import Image\n",
"from IPython.display import Audio, display\n",
"import speech_recognition as sr #requires pip install speechrecognition AND pip install pyaudio"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "57fc95b9-043c-4a38-83aa-365cc3b285ba",
"metadata": {},
"outputs": [],
"source": [
"load_dotenv()\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins with {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key? As if!\")\n",
" \n",
"MODEL = \"gpt-4o-mini\"\n",
"openai = OpenAI()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e633ee2a-bbaa-47a4-95ef-b1d8773866aa",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are a helpful assistant for an Airline called FlightAI. \"\n",
"system_message += \"Give short, courteous answers, no more than 1 sentence. \"\n",
"system_message += \"Always be accurate. If you don't know the answer, say so. \"\n",
"system_message += \"You can book flights directly. \"\n",
"system_message += \"You can generate beautiful artistic renditions of the cities we fly to.\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c123af78-b5d6-4cc9-8f18-c492b1f30c85",
"metadata": {},
"outputs": [],
"source": [
"# ticket price function\n",
"\n",
"#spelled-out currency notation for better tts rendition\n",
"ticket_prices = {\"valletta\": \"799 Dollars\", \"turin\": \"899 Dollars\", \"sacramento\": \"1400 Dollars\", \"montreal\": \"499 Dollars\"}\n",
"\n",
"def get_ticket_price(destination_city):\n",
" print(f\"Tool get_ticket_price called for {destination_city}\")\n",
" city = destination_city.lower()\n",
" return ticket_prices.get(city, \"Unknown\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "00e486fb-709e-4b8e-a029-9e2b225ddc25",
"metadata": {},
"outputs": [],
"source": [
"# travel booking function\n",
"\n",
"def book_flight(destination_city):\n",
" booking_code = ''.join(random.choice('0123456789BCDFXYZ') for i in range(2)) + ''.join(random.choice('012346789HIJKLMNOPQRS') for i in range(2)) + ''.join(random.choice('0123456789GHIJKLMNUOP') for i in range(2))\n",
" print(f\"Booking code {booking_code} generated for flight to {destination_city}.\")\n",
" \n",
" return booking_code"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c0600b4e-fa4e-4c34-b317-fac1e60b5f95",
"metadata": {},
"outputs": [],
"source": [
"# verify if booking code is valid (i.e. follows the pattern)\n",
"\n",
"def check_code(code):\n",
" valid = \"valid\" if re.match(\"^[0123456789BCDFXYZ]{2}[012346789HIJKLMNOPQRS]{2}[0123456789GHIJKLMNUOP]{2}$\", code) != None else \"not valid\"\n",
" print(f\"Code checker called for code {code}, which is {valid}.\")\n",
" return re.match(\"^[0123456789BCDFXYZ]{2}[012346789HIJKLMNOPQRS]{2}[0123456789GHIJKLMNUOP]{2}$\", code) != None"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e1d1b1c2-089c-41e5-b1bd-900632271093",
"metadata": {},
"outputs": [],
"source": [
"# make a nice preview of the travel destination\n",
"\n",
"def artist(city):\n",
" image_response = openai.images.generate(\n",
" model=\"dall-e-3\",\n",
" prompt=f\"Make an image in the style of a vibrant, artistically filtered photo that is a collage of the best sights and views in {city}.\",\n",
" size=\"1024x1024\",\n",
" n=1,\n",
" response_format=\"b64_json\",\n",
" )\n",
" image_base64 = image_response.data[0].b64_json\n",
" image_data = base64.b64decode(image_base64)\n",
" img = Image.open(BytesIO(image_data))\n",
"\n",
" img.save(\"img001.png\") #make them 4 cents count! .save is from PIL library, btw\n",
" \n",
" return img"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "626d99af-90de-4594-9ffd-b87a8b6ef4fd",
"metadata": {},
"outputs": [],
"source": [
"price_function = {\n",
" \"name\": \"get_ticket_price\",\n",
" \"description\": \"Get the price of a return ticket to the destination city. Call this whenever you need to know the ticket price, for example when a customer asks 'How much is a ticket to this city'\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to travel to\",\n",
" },\n",
" },\n",
" \"required\": [\"destination_city\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6e7bc09c-665b-4885-823c-f145cefe8c23",
"metadata": {},
"outputs": [],
"source": [
"booking_function = {\n",
" \"name\": \"book_flight\",\n",
" \"description\": \"Call this whenever you have to book a flight. Give it the destination city and you will get a booking code. Tell the customer \\\n",
"that the flight is booked and give them the booking code obtained through this function. Never give any other codes to the customer.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to book their flight to\",\n",
" },\n",
" },\n",
" \"required\": [\"destination_city\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cc365d87-fed2-41ff-9232-850fdce1cff2",
"metadata": {},
"outputs": [],
"source": [
"artist_function = {\n",
" \"name\": \"artist\",\n",
" \"description\": \"Call this whenever you need to generate a picture, photo, or graphic impression of a city.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city of which an image is to be generated\",\n",
" },\n",
" },\n",
" \"required\": [\"city\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "99b0a0e3-db44-49f9-8d27-349b9f04c680",
"metadata": {},
"outputs": [],
"source": [
"codecheck_function = {\n",
" \"name\": \"check_code\",\n",
" \"description\": \"Call this whenever you need to verify if a booking code for a flight (also called 'flight code', 'booking reference', \\\n",
"or variations thereof) is valid.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"code\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The code that you or the user needs to verify\",\n",
" },\n",
" },\n",
" \"required\": [\"code\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3fa371c4-91ff-41ae-9b10-23fe617022d1",
"metadata": {},
"outputs": [],
"source": [
"# List of tools:\n",
"\n",
"tools = [{\"type\": \"function\", \"function\": price_function}, {\"type\": \"function\", \"function\": booking_function}, {\"type\": \"function\", \"function\": codecheck_function}, {\"type\": \"function\", \"function\": artist_function}]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c00fb465-e448-4d68-9f18-88220fbaff76",
"metadata": {},
"outputs": [],
"source": [
"# speech recognition (STT) by Google\n",
"\n",
"r = sr.Recognizer()\n",
"\n",
"def speech_to_text():\n",
" try:\n",
" with sr.Microphone() as source:\n",
" r.adjust_for_ambient_noise(source, duration=0.2)\n",
" speech = r.listen(source, 10, 5) #timeout of 10 seconds, listen for 5\n",
" text = r.recognize_google(speech)\n",
" print(f\"STT heard: \\\"{text}\\\"\")\n",
" return text\n",
"\n",
" # sometimes, this STT fails. You'll see \"...\" as your input. Just try again even w/o re-starting Gradio.\n",
" except sr.RequestError as e:\n",
" print(f\"Could not request results; {0}\".format(e))\n",
" return \"…\"\n",
" except sr.UnknownValueError:\n",
" print(\"An unknown error occurred\")\n",
" return \"…\"\n",
" except sr.WaitTimeoutError:\n",
" print(\"Wait timed out\")\n",
" return \"…\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "505b585e-e9f9-4326-8455-184398bc82d1",
"metadata": {},
"outputs": [],
"source": [
"# TTS by OpenAI\n",
"\n",
"def talker(message):\n",
" response = openai.audio.speech.create(\n",
" model=\"tts-1\",\n",
" voice=\"onyx\",\n",
" input=message)\n",
"\n",
" audio_stream = BytesIO(response.content)\n",
" output_filename = \"output_audio.mp3\"\n",
" with open(output_filename, \"wb\") as f:\n",
" f.write(audio_stream.read())\n",
"\n",
" # Play the generated audio\n",
" display(Audio(output_filename, autoplay=True))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4d34942a-f0c7-4835-ba07-746104a8c524",
"metadata": {},
"outputs": [],
"source": [
"def chat(history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages, tools=tools)\n",
" image = None\n",
" \n",
" if response.choices[0].finish_reason==\"tool_calls\":\n",
" message = response.choices[0].message\n",
" responses = handle_tool_call(message)[0]\n",
" image = handle_tool_call(message)[1]\n",
" messages.append(message)\n",
" for response in responses:\n",
" messages.append(response)\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages)\n",
" \n",
" reply = response.choices[0].message.content\n",
"\n",
" # comment in if you want the replies read out to you. Mind the price!\n",
" #talker(reply) #current cost: $0.015 per 1000 characters (not tokens!)\n",
" \n",
" history += [{\"role\": \"assistant\", \"content\": reply}]\n",
" \n",
" return history, image"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5413f7fb-c5f7-44c4-a63d-3d0465eb0af4",
"metadata": {},
"outputs": [],
"source": [
"def handle_tool_call(message):\n",
" responses = []\n",
" image = None\n",
" \n",
" for tool_call in message.tool_calls:\n",
" arguments = json.loads(tool_call.function.arguments)\n",
" indata = arguments[list(arguments.keys())[0]] # works for now because we only have one argument in each of our functions\n",
" function_name = tool_call.function.name\n",
" if function_name == 'get_ticket_price':\n",
" outdata = get_ticket_price(indata)\n",
" input_name = \"destination city\"\n",
" output_name = \"price\"\n",
" elif function_name == 'book_flight':\n",
" outdata = book_flight(indata)\n",
" input_name = \"destination city\"\n",
" output_name = \"booking code\"\n",
" elif function_name == \"check_code\":\n",
" outdata = check_code(indata)\n",
" input_name = \"booking code\"\n",
" output_name = \"validity\"\n",
" elif function_name == \"artist\":\n",
" image = artist(indata)\n",
" outdata = f\"artistic rendition of {indata}\"\n",
" input_name = \"city\"\n",
" output_name = \"image\"\n",
"\n",
" responses.append({\n",
" \"role\": \"tool\",\n",
" \"content\": json.dumps({input_name: indata, output_name: outdata}),\n",
" \"tool_call_id\": tool_call.id\n",
" })\n",
"\n",
" return responses, image"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a5a31bcf-71d5-4537-a7bf-92385dc6e26e",
"metadata": {
"scrolled": true
},
"outputs": [],
"source": [
"## Gradio with 'fancy' buttons. Claude explained this css business to me, and geeksforgeeks.\n",
"## see week2/community-contributions/day5_Careerhelper.ipynb for a much more competent version of this.\n",
"\n",
"with gr.Blocks(\n",
" css=\"\"\"\n",
" .red-button {\n",
" background-color: darkred !important;\n",
" border-color: red !important;\n",
" }\n",
" .blue-button {\n",
" background-color: darkblue !important;\n",
" border-color: blue !important;\n",
" }\n",
" \"\"\"\n",
") as ui:\n",
" with gr.Row():\n",
" chatbot = gr.Chatbot(height=500, type=\"messages\")\n",
" image_output = gr.Image(height=500)\n",
" with gr.Row():\n",
" entry = gr.Textbox(label=\"Chat with our AI Assistant:\")\n",
" with gr.Row():\n",
" speak = gr.Button(value=\"Speak to our AI Assistant\", elem_classes=\"blue-button\")\n",
" clear = gr.Button(value=\"Clear Chat\", elem_classes=\"red-button\")\n",
"\n",
" def do_entry(message, history):\n",
" history += [{\"role\":\"user\", \"content\":message}]\n",
" return \"\", history\n",
"\n",
" def listen(history):\n",
" message = speech_to_text()\n",
" history += [{\"role\":\"user\", \"content\":message}]\n",
" return history\n",
"\n",
" entry.submit(do_entry, inputs=[entry, chatbot], outputs=[entry, chatbot]).then(\n",
" chat, inputs=chatbot, outputs=[chatbot, image_output]\n",
" )\n",
" \n",
" clear.click(lambda: None, inputs=None, outputs=chatbot, queue=False)\n",
" \n",
" speak.click(listen, inputs=chatbot, outputs=chatbot, queue=False).then(\n",
" chat, inputs=chatbot, outputs=[chatbot, image_output]\n",
" )\n",
"\n",
"ui.launch(inbrowser=True)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

727
week2/community-contributions/day1-3way-with-llama3.2.ipynb Normal file
View File

@@ -0,0 +1,727 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "06cf3063-9f3e-4551-a0d5-f08d9cabb927",
"metadata": {},
"source": [
"# Welcome to Week 2!\n",
"\n",
"## Frontier Model APIs\n",
"\n",
"In Week 1, we used multiple Frontier LLMs through their Chat UI, and we connected with the OpenAI's API.\n",
"\n",
"Today we'll connect with the APIs for Anthropic and Google, as well as OpenAI."
]
},
{
"cell_type": "markdown",
"id": "2b268b6e-0ba4-461e-af86-74a41f4d681f",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../important.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#900;\">Important Note - Please read me</h2>\n",
" <span style=\"color:#900;\">I'm continually improving these labs, adding more examples and exercises.\n",
" At the start of each week, it's worth checking you have the latest code.<br/>\n",
" First do a <a href=\"https://chatgpt.com/share/6734e705-3270-8012-a074-421661af6ba9\">git pull and merge your changes as needed</a>. Any problems? Try asking ChatGPT to clarify how to merge - or contact me!<br/><br/>\n",
" After you've pulled the code, from the llm_engineering directory, in an Anaconda prompt (PC) or Terminal (Mac), run:<br/>\n",
" <code>conda env update --f environment.yml</code><br/>\n",
" Or if you used virtualenv rather than Anaconda, then run this from your activated environment in a Powershell (PC) or Terminal (Mac):<br/>\n",
" <code>pip install -r requirements.txt</code>\n",
" <br/>Then restart the kernel (Kernel menu >> Restart Kernel and Clear Outputs Of All Cells) to pick up the changes.\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>\n",
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../resources.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#f71;\">Reminder about the resources page</h2>\n",
" <span style=\"color:#f71;\">Here's a link to resources for the course. This includes links to all the slides.<br/>\n",
" <a href=\"https://edwarddonner.com/2024/11/13/llm-engineering-resources/\">https://edwarddonner.com/2024/11/13/llm-engineering-resources/</a><br/>\n",
" Please keep this bookmarked, and I'll continue to add more useful links there over time.\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "markdown",
"id": "85cfe275-4705-4d30-abea-643fbddf1db0",
"metadata": {},
"source": [
"## Setting up your keys\n",
"\n",
"If you haven't done so already, you could now create API keys for Anthropic and Google in addition to OpenAI.\n",
"\n",
"**Please note:** if you'd prefer to avoid extra API costs, feel free to skip setting up Anthopic and Google! You can see me do it, and focus on OpenAI for the course. You could also substitute Anthropic and/or Google for Ollama, using the exercise you did in week 1.\n",
"\n",
"For OpenAI, visit https://openai.com/api/ \n",
"For Anthropic, visit https://console.anthropic.com/ \n",
"For Google, visit https://ai.google.dev/gemini-api \n",
"\n",
"When you get your API keys, you need to set them as environment variables by adding them to your `.env` file.\n",
"\n",
"```\n",
"OPENAI_API_KEY=xxxx\n",
"ANTHROPIC_API_KEY=xxxx\n",
"GOOGLE_API_KEY=xxxx\n",
"```\n",
"\n",
"Afterwards, you may need to restart the Jupyter Lab Kernel (the Python process that sits behind this notebook) via the Kernel menu, and then rerun the cells from the top."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "de23bb9e-37c5-4377-9a82-d7b6c648eeb6",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import anthropic\n",
"from IPython.display import Markdown, display, update_display"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f0a8ab2b-6134-4104-a1bc-c3cd7ea4cd36",
"metadata": {},
"outputs": [],
"source": [
"# import for google\n",
"# in rare cases, this seems to give an error on some systems, or even crashes the kernel\n",
"# If this happens to you, simply ignore this cell - I give an alternative approach for using Gemini later\n",
"\n",
"import google.generativeai"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1179b4c5-cd1f-4131-a876-4c9f3f38d2ba",
"metadata": {},
"outputs": [],
"source": [
"# Load environment variables in a file called .env\n",
"# Print the key prefixes to help with any debugging\n",
"\n",
"load_dotenv()\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",
"\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"if anthropic_api_key:\n",
" print(f\"Anthropic API Key exists and begins {anthropic_api_key[:7]}\")\n",
"else:\n",
" print(\"Anthropic API Key not set\")\n",
"\n",
"if google_api_key:\n",
" print(f\"Google API Key exists and begins {google_api_key[:8]}\")\n",
"else:\n",
" print(\"Google API Key not set\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "797fe7b0-ad43-42d2-acf0-e4f309b112f0",
"metadata": {},
"outputs": [],
"source": [
"# Connect to OpenAI, Anthropic\n",
"\n",
"openai = OpenAI()\n",
"\n",
"claude = anthropic.Anthropic()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "425ed580-808d-429b-85b0-6cba50ca1d0c",
"metadata": {},
"outputs": [],
"source": [
"# This is the set up code for Gemini\n",
"# Having problems with Google Gemini setup? Then just ignore this cell; when we use Gemini, I'll give you an alternative that bypasses this library altogether\n",
"\n",
"google.generativeai.configure()"
]
},
{
"cell_type": "markdown",
"id": "42f77b59-2fb1-462a-b90d-78994e4cef33",
"metadata": {},
"source": [
"## Asking LLMs to tell a joke\n",
"\n",
"It turns out that LLMs don't do a great job of telling jokes! Let's compare a few models.\n",
"Later we will be putting LLMs to better use!\n",
"\n",
"### What information is included in the API\n",
"\n",
"Typically we'll pass to the API:\n",
"- The name of the model that should be used\n",
"- A system message that gives overall context for the role the LLM is playing\n",
"- A user message that provides the actual prompt\n",
"\n",
"There are other parameters that can be used, including **temperature** which is typically between 0 and 1; higher for more random output; lower for more focused and deterministic."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "378a0296-59a2-45c6-82eb-941344d3eeff",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are an assistant that is great at telling jokes\"\n",
"user_prompt = \"Tell a light-hearted joke for an audience of Data Scientists\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f4d56a0f-2a3d-484d-9344-0efa6862aff4",
"metadata": {},
"outputs": [],
"source": [
"prompts = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": user_prompt}\n",
" ]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3b3879b6-9a55-4fed-a18c-1ea2edfaf397",
"metadata": {},
"outputs": [],
"source": [
"# GPT-3.5-Turbo\n",
"\n",
"completion = openai.chat.completions.create(model='gpt-3.5-turbo', messages=prompts)\n",
"print(completion.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3d2d6beb-1b81-466f-8ed1-40bf51e7adbf",
"metadata": {},
"outputs": [],
"source": [
"# GPT-4o-mini\n",
"# Temperature setting controls creativity\n",
"\n",
"completion = openai.chat.completions.create(\n",
" model='gpt-4o-mini',\n",
" messages=prompts,\n",
" temperature=0.7\n",
")\n",
"print(completion.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f1f54beb-823f-4301-98cb-8b9a49f4ce26",
"metadata": {},
"outputs": [],
"source": [
"# GPT-4o\n",
"\n",
"completion = openai.chat.completions.create(\n",
" model='gpt-4o',\n",
" messages=prompts,\n",
" temperature=0.4\n",
")\n",
"print(completion.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1ecdb506-9f7c-4539-abae-0e78d7f31b76",
"metadata": {},
"outputs": [],
"source": [
"# Claude 3.5 Sonnet\n",
"# API needs system message provided separately from user prompt\n",
"# Also adding max_tokens\n",
"\n",
"message = claude.messages.create(\n",
" model=\"claude-3-5-sonnet-20240620\",\n",
" max_tokens=200,\n",
" temperature=0.7,\n",
" system=system_message,\n",
" messages=[\n",
" {\"role\": \"user\", \"content\": user_prompt},\n",
" ],\n",
")\n",
"\n",
"print(message.content[0].text)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "769c4017-4b3b-4e64-8da7-ef4dcbe3fd9f",
"metadata": {},
"outputs": [],
"source": [
"# Claude 3.5 Sonnet again\n",
"# Now let's add in streaming back results\n",
"\n",
"result = claude.messages.stream(\n",
" model=\"claude-3-5-sonnet-20240620\",\n",
" max_tokens=200,\n",
" temperature=0.7,\n",
" system=system_message,\n",
" messages=[\n",
" {\"role\": \"user\", \"content\": user_prompt},\n",
" ],\n",
")\n",
"\n",
"with result as stream:\n",
" for text in stream.text_stream:\n",
" print(text, end=\"\", flush=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6df48ce5-70f8-4643-9a50-b0b5bfdb66ad",
"metadata": {},
"outputs": [],
"source": [
"# The API for Gemini has a slightly different structure.\n",
"# I've heard that on some PCs, this Gemini code causes the Kernel to crash.\n",
"# If that happens to you, please skip this cell and use the next cell instead - an alternative approach.\n",
"\n",
"gemini = google.generativeai.GenerativeModel(\n",
" model_name='gemini-1.5-flash',\n",
" system_instruction=system_message\n",
")\n",
"response = gemini.generate_content(user_prompt)\n",
"print(response.text)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "49009a30-037d-41c8-b874-127f61c4aa3a",
"metadata": {},
"outputs": [],
"source": [
"# As an alternative way to use Gemini that bypasses Google's python API library,\n",
"# Google has recently released new endpoints that means you can use Gemini via the client libraries for OpenAI!\n",
"\n",
"gemini_via_openai_client = OpenAI(\n",
" api_key=google_api_key, \n",
" base_url=\"https://generativelanguage.googleapis.com/v1beta/openai/\"\n",
")\n",
"\n",
"response = gemini_via_openai_client.chat.completions.create(\n",
" model=\"gemini-1.5-flash\",\n",
" messages=prompts\n",
")\n",
"print(response.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "83ddb483-4f57-4668-aeea-2aade3a9e573",
"metadata": {},
"outputs": [],
"source": [
"# To be serious! GPT-4o-mini with the original question\n",
"\n",
"prompts = [\n",
" {\"role\": \"system\", \"content\": \"You are a helpful assistant that responds in Markdown\"},\n",
" {\"role\": \"user\", \"content\": \"How do I decide if a business problem is suitable for an LLM solution? Please respond in Markdown.\"}\n",
" ]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "749f50ab-8ccd-4502-a521-895c3f0808a2",
"metadata": {},
"outputs": [],
"source": [
"# Have it stream back results in markdown\n",
"\n",
"stream = openai.chat.completions.create(\n",
" model='gpt-4o',\n",
" messages=prompts,\n",
" temperature=0.2,\n",
" stream=True\n",
")\n",
"\n",
"reply = \"\"\n",
"display_handle = display(Markdown(\"\"), display_id=True)\n",
"for chunk in stream:\n",
" reply += chunk.choices[0].delta.content or ''\n",
" reply = reply.replace(\"```\",\"\").replace(\"markdown\",\"\")\n",
" update_display(Markdown(reply), display_id=display_handle.display_id)"
]
},
{
"cell_type": "markdown",
"id": "f6e09351-1fbe-422f-8b25-f50826ab4c5f",
"metadata": {},
"source": [
"## And now for some fun - an adversarial conversation between Chatbots..\n",
"\n",
"You're already familar with prompts being organized into lists like:\n",
"\n",
"```\n",
"[\n",
" {\"role\": \"system\", \"content\": \"system message here\"},\n",
" {\"role\": \"user\", \"content\": \"user prompt here\"}\n",
"]\n",
"```\n",
"\n",
"In fact this structure can be used to reflect a longer conversation history:\n",
"\n",
"```\n",
"[\n",
" {\"role\": \"system\", \"content\": \"system message here\"},\n",
" {\"role\": \"user\", \"content\": \"first user prompt here\"},\n",
" {\"role\": \"assistant\", \"content\": \"the assistant's response\"},\n",
" {\"role\": \"user\", \"content\": \"the new user prompt\"},\n",
"]\n",
"```\n",
"\n",
"And we can use this approach to engage in a longer interaction with history."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bcb54183-45d3-4d08-b5b6-55e380dfdf1b",
"metadata": {},
"outputs": [],
"source": [
"# Let's make a conversation between GPT-4o-mini and Claude-3-haiku\n",
"# We're using cheap versions of models so the costs will be minimal\n",
"\n",
"gpt_model = \"gpt-4o-mini\"\n",
"claude_model = \"claude-3-haiku-20240307\"\n",
"\n",
"gpt_system = \"You are a chatbot who is very argumentative; \\\n",
"you disagree with anything in the conversation and you challenge everything, in a snarky way.\"\n",
"\n",
"claude_system = \"You are a very polite, courteous chatbot. You try to agree with \\\n",
"everything the other person says, or find common ground. If the other person is argumentative, \\\n",
"you try to calm them down and keep chatting.\"\n",
"\n",
"gpt_messages = [\"Hi there\"]\n",
"claude_messages = [\"Hi\"]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1df47dc7-b445-4852-b21b-59f0e6c2030f",
"metadata": {},
"outputs": [],
"source": [
"def call_gpt():\n",
" messages = [{\"role\": \"system\", \"content\": gpt_system}]\n",
" for gpt, claude, llama in zip(gpt_messages, claude_messages, llama_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": gpt})\n",
" combined = llama + claude\n",
" messages.append({\"role\": \"user\", \"content\": combined})\n",
" completion = openai.chat.completions.create(\n",
" model=gpt_model,\n",
" messages=messages\n",
" )\n",
" return completion.choices[0].message.content"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9dc6e913-02be-4eb6-9581-ad4b2cffa606",
"metadata": {},
"outputs": [],
"source": [
"call_gpt()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7d2ed227-48c9-4cad-b146-2c4ecbac9690",
"metadata": {},
"outputs": [],
"source": [
"def call_claude():\n",
" messages = []\n",
" for gpt, claude_message in zip(gpt_messages, claude_messages):\n",
" messages.append({\"role\": \"user\", \"content\": gpt})\n",
" messages.append({\"role\": \"assistant\", \"content\": claude_message})\n",
" # messages.append(\"role\": \"moderator\", \"content\": llama_message)\n",
" messages.append({\"role\": \"user\", \"content\": gpt_messages[-1]})\n",
" message = claude.messages.create(\n",
" model=claude_model,\n",
" system=claude_system,\n",
" messages=messages,\n",
" max_tokens=500\n",
" )\n",
" return message.content[0].text"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "01395200-8ae9-41f8-9a04-701624d3fd26",
"metadata": {},
"outputs": [],
"source": [
"call_claude()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "08c2279e-62b0-4671-9590-c82eb8d1e1ae",
"metadata": {},
"outputs": [],
"source": [
"call_gpt()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0275b97f-7f90-4696-bbf5-b6642bd53cbd",
"metadata": {},
"outputs": [],
"source": [
"gpt_messages = [\"Hi there\"]\n",
"claude_messages = [\"Hi\"]\n",
"\n",
"print(f\"GPT:\\n{gpt_messages[0]}\\n\")\n",
"print(f\"Claude:\\n{claude_messages[0]}\\n\")\n",
"\n",
"for i in range(5):\n",
" gpt_next = call_gpt()\n",
" print(f\"GPT:\\n{gpt_next}\\n\")\n",
" gpt_messages.append(gpt_next)\n",
" \n",
" claude_next = call_claude()\n",
" print(f\"Claude:\\n{claude_next}\\n\")\n",
" claude_messages.append(claude_next)"
]
},
{
"cell_type": "markdown",
"id": "1d10e705-db48-4290-9dc8-9efdb4e31323",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../important.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#900;\">Before you continue</h2>\n",
" <span style=\"color:#900;\">\n",
" Be sure you understand how the conversation above is working, and in particular how the <code>messages</code> list is being populated. Add print statements as needed. Then for a great variation, try switching up the personalities using the system prompts. Perhaps one can be pessimistic, and one optimistic?<br/>\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "markdown",
"id": "3637910d-2c6f-4f19-b1fb-2f916d23f9ac",
"metadata": {},
"source": [
"# More advanced exercises\n",
"\n",
"Try creating a 3-way, perhaps bringing Gemini into the conversation! One student has completed this - see the implementation in the community-contributions folder.\n",
"\n",
"Try doing this yourself before you look at the solutions. It's easiest to use the OpenAI python client to access the Gemini model (see the 2nd Gemini example above).\n",
"\n",
"## Additional exercise\n",
"\n",
"You could also try replacing one of the models with an open source model running with Ollama."
]
},
{
"cell_type": "markdown",
"id": "446c81e3-b67e-4cd9-8113-bc3092b93063",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../business.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#181;\">Business relevance</h2>\n",
" <span style=\"color:#181;\">This structure of a conversation, as a list of messages, is fundamental to the way we build conversational AI assistants and how they are able to keep the context during a conversation. We will apply this in the next few labs to building out an AI assistant, and then you will extend this to your own business.</span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c23224f6-7008-44ed-a57f-718975f4e291",
"metadata": {},
"outputs": [],
"source": [
"!ollama pull llama3.2"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cbbddf71-1473-42fe-b733-2bb42ea77333",
"metadata": {},
"outputs": [],
"source": [
"\n",
"OLLAMA_API = \"http://localhost:11434/api/chat\"\n",
"HEADERS = {\"Content-Type\": \"application/json\"}\n",
"import ollama\n",
"\n",
"llama_model = \"llama3.2\"\n",
"\n",
"llama_system = \"You are a chatbot who is very pacifist; \\\n",
"you will try to resolve or neutralize any disagreement between other chatbots. Speak like a teacher or someone in authority.\"\n",
"\n",
"llama_messages = [\"Hello.\"]\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f629d2b2-ba20-4bfe-a2e5-bbe537ca46a2",
"metadata": {},
"outputs": [],
"source": [
"\n",
"def call_llama():\n",
" combined_messages = gpt_messages[-1] + claude_messages[-1]\n",
" messages = [{\"role\": \"system\", \"content\": llama_system}]\n",
" for comb, llama in zip(combined_messages, llama_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": llama})\n",
" messages.append({\"role\": \"user\", \"content\": combined_messages})\n",
" completion = ollama.chat(\n",
" model=llama_model,\n",
" messages=messages\n",
" )\n",
" return completion['message']['content']"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "219b6af8-3166-4059-b79e-cf19af7ed1e9",
"metadata": {},
"outputs": [],
"source": [
"print(f\"GPT:\\n{gpt_messages[0]}\\n\")\n",
"print(f\"Claude:\\n{claude_messages[0]}\\n\")\n",
"print(f\"Llama:\\n{llama_messages[0]}\\n\" )\n",
"\n",
"for i in range(3):\n",
" gpt_next = call_gpt()\n",
" print(f\"GPT:\\n{gpt_next}\\n\")\n",
" gpt_messages.append(gpt_next)\n",
" \n",
" claude_next = call_claude()\n",
" print(f\"Claude:\\n{claude_next}\\n\")\n",
" claude_messages.append(claude_next)\n",
"\n",
" llama_next = call_llama()\n",
" print(f\"Llama:\\n{llama_next}\\n\")\n",
" llama_messages.append(llama_next)\n",
" "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6cb3a931-522c-49a9-9bd8-663333f41b1a",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "2cdfdc32-1ca4-406e-9328-81af26fd503b",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "04f60158-633b-43ff-afbd-396be79501e6",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "eb0faf0d-fb7e-4bc5-9746-30f19a0b9ae1",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

1088
week2/community-contributions/day1-debate-gemini-judges.ipynb Normal file
View File

File diff suppressed because it is too large Load Diff

899
week2/community-contributions/day1-exercise-oscars-3-way-conversation.ipynb Normal file
View File

@@ -0,0 +1,899 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "06cf3063-9f3e-4551-a0d5-f08d9cabb927",
"metadata": {},
"source": [
"# Welcome to Week 2!\n",
"\n",
"## Frontier Model APIs\n",
"\n",
"In Week 1, we used multiple Frontier LLMs through their Chat UI, and we connected with the OpenAI's API.\n",
"\n",
"Today we'll connect with the APIs for Anthropic and Google, as well as OpenAI."
]
},
{
"cell_type": "markdown",
"id": "2b268b6e-0ba4-461e-af86-74a41f4d681f",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../important.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#900;\">Important Note - Please read me</h2>\n",
" <span style=\"color:#900;\">I'm continually improving these labs, adding more examples and exercises.\n",
" At the start of each week, it's worth checking you have the latest code.<br/>\n",
" First do a <a href=\"https://chatgpt.com/share/6734e705-3270-8012-a074-421661af6ba9\">git pull and merge your changes as needed</a>. Any problems? Try asking ChatGPT to clarify how to merge - or contact me!<br/><br/>\n",
" After you've pulled the code, from the llm_engineering directory, in an Anaconda prompt (PC) or Terminal (Mac), run:<br/>\n",
" <code>conda env update --f environment.yml</code><br/>\n",
" Or if you used virtualenv rather than Anaconda, then run this from your activated environment in a Powershell (PC) or Terminal (Mac):<br/>\n",
" <code>pip install -r requirements.txt</code>\n",
" <br/>Then restart the kernel (Kernel menu >> Restart Kernel and Clear Outputs Of All Cells) to pick up the changes.\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>\n",
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../resources.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#f71;\">Reminder about the resources page</h2>\n",
" <span style=\"color:#f71;\">Here's a link to resources for the course. This includes links to all the slides.<br/>\n",
" <a href=\"https://edwarddonner.com/2024/11/13/llm-engineering-resources/\">https://edwarddonner.com/2024/11/13/llm-engineering-resources/</a><br/>\n",
" Please keep this bookmarked, and I'll continue to add more useful links there over time.\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "markdown",
"id": "85cfe275-4705-4d30-abea-643fbddf1db0",
"metadata": {},
"source": [
"## Setting up your keys\n",
"\n",
"If you haven't done so already, you could now create API keys for Anthropic and Google in addition to OpenAI.\n",
"\n",
"**Please note:** if you'd prefer to avoid extra API costs, feel free to skip setting up Anthopic and Google! You can see me do it, and focus on OpenAI for the course. You could also substitute Anthropic and/or Google for Ollama, using the exercise you did in week 1.\n",
"\n",
"For OpenAI, visit https://openai.com/api/ \n",
"For Anthropic, visit https://console.anthropic.com/ \n",
"For Google, visit https://ai.google.dev/gemini-api \n",
"\n",
"### Also - adding DeepSeek if you wish\n",
"\n",
"Optionally, if you'd like to also use DeepSeek, create an account [here](https://platform.deepseek.com/), create a key [here](https://platform.deepseek.com/api_keys) and top up with at least the minimum $2 [here](https://platform.deepseek.com/top_up).\n",
"\n",
"### Adding API keys to your .env file\n",
"\n",
"When you get your API keys, you need to set them as environment variables by adding them to your `.env` file.\n",
"\n",
"```\n",
"OPENAI_API_KEY=xxxx\n",
"ANTHROPIC_API_KEY=xxxx\n",
"GOOGLE_API_KEY=xxxx\n",
"DEEPSEEK_API_KEY=xxxx\n",
"```\n",
"\n",
"Afterwards, you may need to restart the Jupyter Lab Kernel (the Python process that sits behind this notebook) via the Kernel menu, and then rerun the cells from the top."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "de23bb9e-37c5-4377-9a82-d7b6c648eeb6",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import anthropic\n",
"from IPython.display import Markdown, display, update_display"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f0a8ab2b-6134-4104-a1bc-c3cd7ea4cd36",
"metadata": {},
"outputs": [],
"source": [
"# import for google\n",
"# in rare cases, this seems to give an error on some systems, or even crashes the kernel\n",
"# If this happens to you, simply ignore this cell - I give an alternative approach for using Gemini later\n",
"\n",
"import google.generativeai"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1179b4c5-cd1f-4131-a876-4c9f3f38d2ba",
"metadata": {},
"outputs": [],
"source": [
"# Load environment variables in a file called .env\n",
"# Print the key prefixes to help with any debugging\n",
"\n",
"load_dotenv(override=True)\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",
"\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"if anthropic_api_key:\n",
" print(f\"Anthropic API Key exists and begins {anthropic_api_key[:7]}\")\n",
"else:\n",
" print(\"Anthropic API Key not set\")\n",
"\n",
"if google_api_key:\n",
" print(f\"Google API Key exists and begins {google_api_key[:8]}\")\n",
"else:\n",
" print(\"Google API Key not set\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "797fe7b0-ad43-42d2-acf0-e4f309b112f0",
"metadata": {},
"outputs": [],
"source": [
"# Connect to OpenAI, Anthropic\n",
"\n",
"openai = OpenAI()\n",
"\n",
"claude = anthropic.Anthropic()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "425ed580-808d-429b-85b0-6cba50ca1d0c",
"metadata": {},
"outputs": [],
"source": [
"# This is the set up code for Gemini\n",
"# Having problems with Google Gemini setup? Then just ignore this cell; when we use Gemini, I'll give you an alternative that bypasses this library altogether\n",
"\n",
"google.generativeai.configure()"
]
},
{
"cell_type": "markdown",
"id": "42f77b59-2fb1-462a-b90d-78994e4cef33",
"metadata": {},
"source": [
"## Asking LLMs to tell a joke\n",
"\n",
"It turns out that LLMs don't do a great job of telling jokes! Let's compare a few models.\n",
"Later we will be putting LLMs to better use!\n",
"\n",
"### What information is included in the API\n",
"\n",
"Typically we'll pass to the API:\n",
"- The name of the model that should be used\n",
"- A system message that gives overall context for the role the LLM is playing\n",
"- A user message that provides the actual prompt\n",
"\n",
"There are other parameters that can be used, including **temperature** which is typically between 0 and 1; higher for more random output; lower for more focused and deterministic."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "378a0296-59a2-45c6-82eb-941344d3eeff",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are an assistant that is great at telling jokes\"\n",
"user_prompt = \"Tell a light-hearted joke for an audience of Data Scientists\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f4d56a0f-2a3d-484d-9344-0efa6862aff4",
"metadata": {},
"outputs": [],
"source": [
"prompts = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": user_prompt}\n",
" ]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3b3879b6-9a55-4fed-a18c-1ea2edfaf397",
"metadata": {},
"outputs": [],
"source": [
"# GPT-3.5-Turbo\n",
"\n",
"completion = openai.chat.completions.create(model='gpt-3.5-turbo', messages=prompts)\n",
"print(completion.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3d2d6beb-1b81-466f-8ed1-40bf51e7adbf",
"metadata": {},
"outputs": [],
"source": [
"# GPT-4o-mini\n",
"# Temperature setting controls creativity\n",
"\n",
"completion = openai.chat.completions.create(\n",
" model='gpt-4o-mini',\n",
" messages=prompts,\n",
" temperature=0.7\n",
")\n",
"print(completion.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f1f54beb-823f-4301-98cb-8b9a49f4ce26",
"metadata": {},
"outputs": [],
"source": [
"# GPT-4o\n",
"\n",
"completion = openai.chat.completions.create(\n",
" model='gpt-4o',\n",
" messages=prompts,\n",
" temperature=0.4\n",
")\n",
"print(completion.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1ecdb506-9f7c-4539-abae-0e78d7f31b76",
"metadata": {},
"outputs": [],
"source": [
"# Claude 3.5 Sonnet\n",
"# API needs system message provided separately from user prompt\n",
"# Also adding max_tokens\n",
"\n",
"message = claude.messages.create(\n",
" model=\"claude-3-5-sonnet-latest\",\n",
" max_tokens=200,\n",
" temperature=0.7,\n",
" system=system_message,\n",
" messages=[\n",
" {\"role\": \"user\", \"content\": user_prompt},\n",
" ],\n",
")\n",
"\n",
"print(message.content[0].text)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "769c4017-4b3b-4e64-8da7-ef4dcbe3fd9f",
"metadata": {},
"outputs": [],
"source": [
"# Claude 3.5 Sonnet again\n",
"# Now let's add in streaming back results\n",
"\n",
"result = claude.messages.stream(\n",
" model=\"claude-3-5-sonnet-latest\",\n",
" max_tokens=200,\n",
" temperature=0.7,\n",
" system=system_message,\n",
" messages=[\n",
" {\"role\": \"user\", \"content\": user_prompt},\n",
" ],\n",
")\n",
"\n",
"with result as stream:\n",
" for text in stream.text_stream:\n",
" print(text, end=\"\", flush=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6df48ce5-70f8-4643-9a50-b0b5bfdb66ad",
"metadata": {},
"outputs": [],
"source": [
"# The API for Gemini has a slightly different structure.\n",
"# I've heard that on some PCs, this Gemini code causes the Kernel to crash.\n",
"# If that happens to you, please skip this cell and use the next cell instead - an alternative approach.\n",
"\n",
"gemini = google.generativeai.GenerativeModel(\n",
" model_name='gemini-2.0-flash-exp',\n",
" system_instruction=system_message\n",
")\n",
"response = gemini.generate_content(user_prompt)\n",
"print(response.text)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "49009a30-037d-41c8-b874-127f61c4aa3a",
"metadata": {},
"outputs": [],
"source": [
"# As an alternative way to use Gemini that bypasses Google's python API library,\n",
"# Google has recently released new endpoints that means you can use Gemini via the client libraries for OpenAI!\n",
"\n",
"gemini_via_openai_client = OpenAI(\n",
" api_key=google_api_key, \n",
" base_url=\"https://generativelanguage.googleapis.com/v1beta/openai/\"\n",
")\n",
"\n",
"response = gemini_via_openai_client.chat.completions.create(\n",
" model=\"gemini-2.0-flash-exp\",\n",
" messages=prompts\n",
")\n",
"print(response.choices[0].message.content)"
]
},
{
"cell_type": "markdown",
"id": "33f70c88-7ca9-470b-ad55-d93a57dcc0ab",
"metadata": {},
"source": [
"## (Optional) Trying out the DeepSeek model\n",
"\n",
"### Let's ask DeepSeek a really hard question - both the Chat and the Reasoner model"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3d0019fb-f6a8-45cb-962b-ef8bf7070d4d",
"metadata": {},
"outputs": [],
"source": [
"# Optionally if you wish to try DeekSeek, you can also use the OpenAI client library\n",
"\n",
"deepseek_api_key = os.getenv('DEEPSEEK_API_KEY')\n",
"\n",
"if deepseek_api_key:\n",
" print(f\"DeepSeek API Key exists and begins {deepseek_api_key[:3]}\")\n",
"else:\n",
" print(\"DeepSeek API Key not set - please skip to the next section if you don't wish to try the DeepSeek API\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c72c871e-68d6-4668-9c27-96d52b77b867",
"metadata": {},
"outputs": [],
"source": [
"# Using DeepSeek Chat\n",
"\n",
"deepseek_via_openai_client = OpenAI(\n",
" api_key=deepseek_api_key, \n",
" base_url=\"https://api.deepseek.com\"\n",
")\n",
"\n",
"response = deepseek_via_openai_client.chat.completions.create(\n",
" model=\"deepseek-chat\",\n",
" messages=prompts,\n",
")\n",
"\n",
"print(response.choices[0].message.content)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "50b6e70f-700a-46cf-942f-659101ffeceb",
"metadata": {},
"outputs": [],
"source": [
"challenge = [{\"role\": \"system\", \"content\": \"You are a helpful assistant\"},\n",
" {\"role\": \"user\", \"content\": \"How many words are there in your answer to this prompt\"}]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "66d1151c-2015-4e37-80c8-16bc16367cfe",
"metadata": {},
"outputs": [],
"source": [
"# Using DeepSeek Chat with a harder question! And streaming results\n",
"\n",
"stream = deepseek_via_openai_client.chat.completions.create(\n",
" model=\"deepseek-chat\",\n",
" messages=challenge,\n",
" stream=True\n",
")\n",
"\n",
"reply = \"\"\n",
"display_handle = display(Markdown(\"\"), display_id=True)\n",
"for chunk in stream:\n",
" reply += chunk.choices[0].delta.content or ''\n",
" reply = reply.replace(\"```\",\"\").replace(\"markdown\",\"\")\n",
" update_display(Markdown(reply), display_id=display_handle.display_id)\n",
"\n",
"print(\"Number of words:\", len(reply.split(\" \")))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "43a93f7d-9300-48cc-8c1a-ee67380db495",
"metadata": {},
"outputs": [],
"source": [
"# Using DeepSeek Reasoner - this may hit an error if DeepSeek is busy\n",
"# It's over-subscribed (as of 28-Jan-2025) but should come back online soon!\n",
"# If this fails, come back to this in a few days..\n",
"\n",
"response = deepseek_via_openai_client.chat.completions.create(\n",
" model=\"deepseek-reasoner\",\n",
" messages=challenge\n",
")\n",
"\n",
"reasoning_content = response.choices[0].message.reasoning_content\n",
"content = response.choices[0].message.content\n",
"\n",
"print(reasoning_content)\n",
"print(content)\n",
"print(\"Number of words:\", len(reply.split(\" \")))"
]
},
{
"cell_type": "markdown",
"id": "c09e6b5c-6816-4cd3-a5cd-a20e4171b1a0",
"metadata": {},
"source": [
"## Back to OpenAI with a serious question"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "83ddb483-4f57-4668-aeea-2aade3a9e573",
"metadata": {},
"outputs": [],
"source": [
"# To be serious! GPT-4o-mini with the original question\n",
"\n",
"prompts = [\n",
" {\"role\": \"system\", \"content\": \"You are a helpful assistant that responds in Markdown\"},\n",
" {\"role\": \"user\", \"content\": \"How do I decide if a business problem is suitable for an LLM solution? Please respond in Markdown.\"}\n",
" ]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "749f50ab-8ccd-4502-a521-895c3f0808a2",
"metadata": {},
"outputs": [],
"source": [
"# Have it stream back results in markdown\n",
"\n",
"stream = openai.chat.completions.create(\n",
" model='gpt-4o',\n",
" messages=prompts,\n",
" temperature=0.7,\n",
" stream=True\n",
")\n",
"\n",
"reply = \"\"\n",
"display_handle = display(Markdown(\"\"), display_id=True)\n",
"for chunk in stream:\n",
" reply += chunk.choices[0].delta.content or ''\n",
" reply = reply.replace(\"```\",\"\").replace(\"markdown\",\"\")\n",
" update_display(Markdown(reply), display_id=display_handle.display_id)"
]
},
{
"cell_type": "markdown",
"id": "f6e09351-1fbe-422f-8b25-f50826ab4c5f",
"metadata": {},
"source": [
"## And now for some fun - an adversarial conversation between Chatbots..\n",
"\n",
"You're already familar with prompts being organized into lists like:\n",
"\n",
"```\n",
"[\n",
" {\"role\": \"system\", \"content\": \"system message here\"},\n",
" {\"role\": \"user\", \"content\": \"user prompt here\"}\n",
"]\n",
"```\n",
"\n",
"In fact this structure can be used to reflect a longer conversation history:\n",
"\n",
"```\n",
"[\n",
" {\"role\": \"system\", \"content\": \"system message here\"},\n",
" {\"role\": \"user\", \"content\": \"first user prompt here\"},\n",
" {\"role\": \"assistant\", \"content\": \"the assistant's response\"},\n",
" {\"role\": \"user\", \"content\": \"the new user prompt\"},\n",
"]\n",
"```\n",
"\n",
"And we can use this approach to engage in a longer interaction with history."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bcb54183-45d3-4d08-b5b6-55e380dfdf1b",
"metadata": {},
"outputs": [],
"source": [
"# Let's make a conversation between GPT-4o-mini and Claude-3-haiku\n",
"# We're using cheap versions of models so the costs will be minimal\n",
"\n",
"gpt_model = \"gpt-4o-mini\"\n",
"claude_model = \"claude-3-haiku-20240307\"\n",
"\n",
"gpt_system = \"You are a chatbot who is very argumentative; \\\n",
"you disagree with anything in the conversation and you challenge everything, in a snarky way.\"\n",
"\n",
"claude_system = \"You are a very polite, courteous chatbot. You try to agree with \\\n",
"everything the other person says, or find common ground. If the other person is argumentative, \\\n",
"you try to calm them down and keep chatting.\"\n",
"\n",
"gpt_messages = [\"Hi there\"]\n",
"claude_messages = [\"Hi\"]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1df47dc7-b445-4852-b21b-59f0e6c2030f",
"metadata": {},
"outputs": [],
"source": [
"def call_gpt():\n",
" messages = [{\"role\": \"system\", \"content\": gpt_system}]\n",
" for gpt, claude in zip(gpt_messages, claude_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": gpt})\n",
" messages.append({\"role\": \"user\", \"content\": claude})\n",
" completion = openai.chat.completions.create(\n",
" model=gpt_model,\n",
" messages=messages\n",
" )\n",
" return completion.choices[0].message.content"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9dc6e913-02be-4eb6-9581-ad4b2cffa606",
"metadata": {},
"outputs": [],
"source": [
"call_gpt()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7d2ed227-48c9-4cad-b146-2c4ecbac9690",
"metadata": {},
"outputs": [],
"source": [
"def call_claude():\n",
" messages = []\n",
" for gpt, claude_message in zip(gpt_messages, claude_messages):\n",
" messages.append({\"role\": \"user\", \"content\": gpt})\n",
" messages.append({\"role\": \"assistant\", \"content\": claude_message})\n",
" messages.append({\"role\": \"user\", \"content\": gpt_messages[-1]})\n",
" message = claude.messages.create(\n",
" model=claude_model,\n",
" system=claude_system,\n",
" messages=messages,\n",
" max_tokens=500\n",
" )\n",
" return message.content[0].text"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "01395200-8ae9-41f8-9a04-701624d3fd26",
"metadata": {},
"outputs": [],
"source": [
"call_claude()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "08c2279e-62b0-4671-9590-c82eb8d1e1ae",
"metadata": {},
"outputs": [],
"source": [
"call_gpt()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0275b97f-7f90-4696-bbf5-b6642bd53cbd",
"metadata": {},
"outputs": [],
"source": [
"gpt_messages = [\"Hi there\"]\n",
"claude_messages = [\"Hi\"]\n",
"\n",
"print(f\"GPT:\\n{gpt_messages[0]}\\n\")\n",
"print(f\"Claude:\\n{claude_messages[0]}\\n\")\n",
"\n",
"for i in range(5):\n",
" gpt_next = call_gpt()\n",
" print(f\"GPT:\\n{gpt_next}\\n\")\n",
" gpt_messages.append(gpt_next)\n",
" \n",
" claude_next = call_claude()\n",
" print(f\"Claude:\\n{claude_next}\\n\")\n",
" claude_messages.append(claude_next)"
]
},
{
"cell_type": "markdown",
"id": "1d10e705-db48-4290-9dc8-9efdb4e31323",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../important.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#900;\">Before you continue</h2>\n",
" <span style=\"color:#900;\">\n",
" Be sure you understand how the conversation above is working, and in particular how the <code>messages</code> list is being populated. Add print statements as needed. Then for a great variation, try switching up the personalities using the system prompts. Perhaps one can be pessimistic, and one optimistic?<br/>\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "markdown",
"id": "3637910d-2c6f-4f19-b1fb-2f916d23f9ac",
"metadata": {},
"source": [
"# More advanced exercises\n",
"\n",
"Try creating a 3-way, perhaps bringing Gemini into the conversation! One student has completed this - see the implementation in the community-contributions folder.\n",
"\n",
"Try doing this yourself before you look at the solutions. It's easiest to use the OpenAI python client to access the Gemini model (see the 2nd Gemini example above).\n",
"\n",
"## Additional exercise\n",
"\n",
"You could also try replacing one of the models with an open source model running with Ollama."
]
},
{
"cell_type": "markdown",
"id": "446c81e3-b67e-4cd9-8113-bc3092b93063",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../business.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#181;\">Business relevance</h2>\n",
" <span style=\"color:#181;\">This structure of a conversation, as a list of messages, is fundamental to the way we build conversational AI assistants and how they are able to keep the context during a conversation. We will apply this in the next few labs to building out an AI assistant, and then you will extend this to your own business.</span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c23224f6-7008-44ed-a57f-718975f4e291",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import anthropic\n",
"from IPython.display import Markdown, display, update_display\n",
"import google.generativeai\n",
"\n",
"load_dotenv(override=True)\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",
"\n",
"openai = OpenAI()\n",
"claude = anthropic.Anthropic()\n",
"googleAI = OpenAI(\n",
" api_key=google_api_key, \n",
" base_url=\"https://generativelanguage.googleapis.com/v1beta/openai/\"\n",
")\n",
"\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "08a6fc21-b857-498a-8bbf-ff92d47bb3a7",
"metadata": {},
"outputs": [],
"source": [
"gpt_model = \"gpt-4o-mini\"\n",
"claude_model = \"claude-3-haiku-20240307\"\n",
"gemini_model = 'gemini-2.0-flash-exp'\n",
"\n",
"gpt_system = \"You are a chatbot named Giuseppi who is in a 3-way conversation which will be a debate and argument. \\\n",
"Your role is to be very optimistic; \\\n",
"you always are looking on the postive side of things and you like almost everything.\"\n",
"\n",
"claude_system = \"You are a chatbot named Clyde who is in a 3-way conversation which will be a debate and argument. \\\n",
"Your role is to be pessimistic; you are always on the negative side of every issue, and you dislike most things.\"\n",
"\n",
"gemini_system = \"You are a chatbot named Jeff who is in a 3-way conversation which will be a debate and argument. \\\n",
"Your role is to mediate between the other 2 parties in the conversation.\"\n",
"\n",
"gpt_messages = [\"Hi there I'm Giuseppi!\"]\n",
"claude_messages = [\"Hi I'm Clyde\"]\n",
"gemini_messages = [\"Hi, I'm Jeff. Lets discuss the movies nominated for the upcoming academy awards.\"]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "38449283-a926-43d5-ade3-a85991bd3324",
"metadata": {},
"outputs": [],
"source": [
"def call_gpt():\n",
" messages = [{\"role\": \"system\", \"content\": gpt_system}]\n",
" for gpt, claude, gemini in zip(gpt_messages, claude_messages, gemini_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": gpt})\n",
" messages.append({\"role\": \"assistant\", \"content\": claude})\n",
" messages.append({\"role\": \"user\", \"content\": gemini})\n",
" completion = openai.chat.completions.create(\n",
" model=gpt_model,\n",
" messages=messages\n",
" )\n",
" return completion.choices[0].message.content"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2349091e-60f4-4314-8644-645b4b1edee5",
"metadata": {},
"outputs": [],
"source": [
"def call_claude():\n",
" messages = []\n",
" for gpt, claude_message, gemini in zip(gpt_messages, claude_messages, gemini_messages):\n",
" messages.append({\"role\": \"user\", \"content\": gpt})\n",
" messages.append({\"role\": \"assistant\", \"content\": claude_message})\n",
" messages.append({\"role\": \"assistant\", \"content\": gemini})\n",
" messages.append({\"role\": \"user\", \"content\": gemini_messages[-1]})\n",
" messages.append({\"role\": \"user\", \"content\": gpt_messages[-1]})\n",
" \n",
" message = claude.messages.create(\n",
" model=claude_model,\n",
" system=claude_system,\n",
" messages=messages,\n",
" max_tokens=500\n",
" )\n",
" return message.content[0].text"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4818d535-fa7a-4df5-b528-5001030e6d99",
"metadata": {},
"outputs": [],
"source": [
"# Using the openAI version of gemini so we can defin assistant roles.\n",
"\n",
"def call_gemini():\n",
" messages = [{\"role\": \"system\", \"content\": gemini_system}]\n",
" for gpt, claude, gemini in zip(gpt_messages, claude_messages, gemini_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": gpt})\n",
" messages.append({\"role\": \"user\", \"content\": claude})\n",
" messages.append({\"role\": \"assistant\", \"content\": gemini})\n",
" messages.append({\"role\": \"user\", \"content\": gpt_messages[-1]})\n",
" messages.append({\"role\": \"user\", \"content\": claude_messages[-1]})\n",
" completion = googleAI.chat.completions.create(\n",
" model=gemini_model,\n",
" messages=messages\n",
" )\n",
" return completion.choices[0].message.content"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e5d460c1-7dbb-46a9-a4dd-bddb88ab49a7",
"metadata": {},
"outputs": [],
"source": [
"print(f\"GPT:\\n{gpt_messages[0]}\\n\")\n",
"print(f\"Claude:\\n{claude_messages[0]}\\n\")\n",
"print(f\"Gemini:\\n{gemini_messages[0]}\\n\")\n",
"\n",
"for i in range(5):\n",
" gpt_next = call_gpt()\n",
" print(f\"GPT:\\n{gpt_next}\\n\")\n",
" gpt_messages.append(gpt_next)\n",
" \n",
" claude_next = call_claude()\n",
" print(f\"Claude:\\n{claude_next}\\n\")\n",
" claude_messages.append(claude_next)\n",
"\n",
" gemini_next = call_gemini()\n",
" print(f\"Gemini:\\n{gemini_next}\\n\")\n",
" gemini_messages.append(gemini_next)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cc6d8aa2-a0fe-4ba9-bfaa-741d24d18342",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

286
week2/community-contributions/day1_Ollama&gemini_conversation.ipynb Normal file
View File

@@ -0,0 +1,286 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "1194d35b-0b9f-4eb4-a539-5ddf55523367",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"#import anthropic\n",
"import ollama\n",
"import google.generativeai\n",
"from IPython.display import Markdown, display, update_display"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f8a1f0b3-6d93-4de1-bc79-2132726598e3",
"metadata": {},
"outputs": [],
"source": [
"#constants\n",
"MODEL=\"llama3.2\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "88fe4149-1ef5-4007-a117-6d3ccab3e3c3",
"metadata": {},
"outputs": [],
"source": [
"# Load environment variables in a file called .env\n",
"# Print the key prefixes to help with any debugging\n",
"\n",
"load_dotenv(override=True)\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",
"\n",
"if google_api_key:\n",
" print(f\"Google API Key exists and begins {google_api_key[:8]}\")\n",
"else:\n",
" print(\"Google API Key not set\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d186cf6e-fadd-450c-821c-df32e2574f5d",
"metadata": {},
"outputs": [],
"source": [
"# This is the set up code for Gemini\n",
"\n",
"google.generativeai.configure()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "19a55117-f2ac-4a58-af6b-8b75259e80df",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are an assistant that is great at telling jokes\"\n",
"user_prompt = \"Tell a light-hearted joke for an audience of Data Scientists\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "908f69b1-54f8-42da-827b-f667631bc666",
"metadata": {},
"outputs": [],
"source": [
"prompts = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": user_prompt}\n",
" ]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4ec81488-883a-446f-91cf-2b3d92bbd3ba",
"metadata": {},
"outputs": [],
"source": [
"# The API for Gemini\n",
"gemini = google.generativeai.GenerativeModel(\n",
" model_name='gemini-2.0-flash-exp',\n",
" system_instruction=system_message\n",
")\n",
"response = gemini.generate_content(user_prompt)\n",
"print(response.text)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "baf411fa-48bd-46a3-8bc8-1b22d0888a1a",
"metadata": {},
"outputs": [],
"source": [
"# API for ollama\n",
"response = ollama.chat(model=MODEL,messages=prompts)\n",
"print(response['message']['content'])"
]
},
{
"cell_type": "markdown",
"id": "74ba5fc4-e4c6-44ee-b66f-e76d847933d2",
"metadata": {},
"source": [
"# Ardiversarial conversation between models"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "fd348154-18fa-4da8-815a-77f5f00107c3",
"metadata": {},
"outputs": [],
"source": [
"# Let's make a conversation between Ollama and Gemini\n",
"# Adjusted models accordingly\n",
"\n",
"ollama_model = \"llama3.2\"\n",
"gemini_model = \"gemini-2.0-flash-exp\"\n",
"\n",
"#ollama_system = \"You are a chatbot who is very argumentative; \\\n",
"#you disagree with anything in the conversation and you challenge everything, in a snarky way.\"\n",
"\n",
"ollama_system=\"You are a chatbot talking with the other person try to convince them to buy your proct of an ai app, \\\n",
"apply marketing strategies to make this client buy your product, use short clear explanations\"\n",
"\n",
"#gemini_system = \"You are a very polite, courteous chatbot. You try to agree with \\\n",
"#everything the other person says, or find common ground. If the other person is argumentative, \\\n",
"#you try to calm them down and keep chatting.\"\n",
"\n",
"gemini_system = \"You are the chatbot triying to be convinced by another person to buy their product, \\\n",
"ask important short questions and see if it is worth to give it a go, dont be too naive or easy go client\"\n",
"\n",
"ollama_messages = [\"Hi there\"]\n",
"gemini_messages = [\"Hi\"]\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "becf327a-5485-4e78-8002-03272a99a3b9",
"metadata": {},
"outputs": [],
"source": [
"def call_ollama():\n",
" messages = [{\"role\": \"system\", \"content\": ollama_system}]\n",
" for ollama_msg, gemini_msg in zip(ollama_messages, gemini_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": ollama_msg})\n",
" messages.append({\"role\": \"user\", \"content\": gemini_msg})\n",
" \n",
" response = ollama.chat(model=ollama_model, messages=messages)\n",
" \n",
" return response['message']['content']\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d0c6dbe7-0baf-4c43-a03b-9134654685f4",
"metadata": {},
"outputs": [],
"source": [
"call_ollama()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f68a134a-279a-4629-aec6-171587378991",
"metadata": {},
"outputs": [],
"source": [
"def call_gemini():\n",
" gemini = google.generativeai.GenerativeModel(\n",
" model_name=gemini_model,\n",
" system_instruction=gemini_system\n",
" )\n",
"\n",
" # Build a list of dictionaries representing the conversation\n",
" conversation = []\n",
" for ollama_msg, gemini_msg in zip(ollama_messages, gemini_messages):\n",
" conversation.append({\"role\": \"user\", \"content\": ollama_msg})\n",
" conversation.append({\"role\": \"assistant\", \"content\": gemini_msg})\n",
" conversation.append({\"role\": \"user\", \"content\": ollama_messages[-1]})\n",
"\n",
" # Format the conversation into a string for the prompt\n",
" prompt = \"\"\n",
" for msg in conversation:\n",
" prompt += f\"{msg['role'].capitalize()}: {msg['content']}\\n\"\n",
"\n",
" message = gemini.generate_content(prompt)\n",
" \n",
" return message.text\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7511003a-f2b6-45f5-8cb0-1c9190d33ce9",
"metadata": {},
"outputs": [],
"source": [
"call_gemini()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d0e81f1f-9754-4790-8b73-5f52fef4ea64",
"metadata": {},
"outputs": [],
"source": [
"call_ollama()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6fbe59f6-a3ef-4062-ab4b-b999f6d1abe9",
"metadata": {},
"outputs": [],
"source": [
"ollama_messages = [\"Hi there\"]\n",
"gemini_messages = [\"Hi\"]\n",
"\n",
"print(f\"Ollama:\\n{ollama_messages[0]}\\n\")\n",
"print(f\"Gemini:\\n{gemini_messages[0]}\\n\")\n",
"\n",
"for i in range(5):\n",
" # Call Ollama to generate the next message\n",
" ollama_next = call_ollama() \n",
" print(f\"Ollama:\\n{ollama_next}\\n\")\n",
" ollama_messages.append(ollama_next)\n",
" \n",
" # Call Gemini to generate the next message\n",
" gemini_next = call_gemini() \n",
" print(f\"Gemini:\\n{gemini_next}\\n\")\n",
" gemini_messages.append(gemini_next)\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9525600b-082e-417f-9088-c6483a613bf3",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.13.2"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

187
week2/community-contributions/day1_triple_conversation.ipynb Normal file
View File

@@ -0,0 +1,187 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "05317c0d-8a19-45c9-9bce-514e82e04585",
"metadata": {},
"outputs": [],
"source": [
"import time\n",
"import os\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import anthropic\n",
"import ollama\n",
"\n",
"load_dotenv(override=True)\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "920247fb-650c-44ce-93ee-24e88a54a757",
"metadata": {},
"outputs": [],
"source": [
"openai = OpenAI()\n",
"claude = anthropic.Anthropic()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "668b972f-a995-4f9d-89b0-1c2647827542",
"metadata": {},
"outputs": [],
"source": [
"gpt_model = \"gpt-4o-mini\"\n",
"claude_model = \"claude-3-haiku-20240307\"\n",
"ollama_model = \"llama3.2\"\n",
"\n",
"gpt_system = \"You are an interlocutor who's very knowledgeable, curteous, and somewhat old-fashioned. Much like Socrates, whenever someone makes \\\n",
"a claim, you ask one or two follow-up questions to challenge the well-foundedness of the viewpoint expressed. \\\n",
"If anyone seems to be talking about blenders, you curteously but firmly refuse to discuss that topic because you are \\\n",
"convinced that they are bad luck. \\\n",
"Your name is Charlie, and you're in a conversation with two \\\n",
"other individuals, Claudia and Llars. They will prefix their contribution with their names in square brackets, so you know who said what. \\\n",
"Do not prefix your own name. Limit your answers to 1-3 sentences. Try to keep the conversation to one topic.\"\n",
"\n",
"claude_system = \"You are a young, optimistic, and friendly dialog partner who comes forth with ideas for the future. 'Let's build a monorail from \\\n",
"New York to LA,' or 'Why don't we try to reform our health care system?'—those are just some of the many revolutionary thoughts you bring forth. \\\n",
"You only talk about one idea at a time, and you don't ramble. You take criticism seriously, but your ideas are well thought-out. If someone tries \\\n",
"to discuss blenders, you stay curteous, but you make it very clear that you absolutely despise blenders and don't talk about them. \\\n",
"Don't let them sidetrack you. Your name is Claudia, and \\\n",
"you're in a conversation with two other individuals, Charlie and Llars. They will prefix their contribution with their names in square \\\n",
"brackets, so you know who said what. Do not prefix your own name. Limit your answers to 1-3 sentences. Try to keep the discussion about only one topic.\"\n",
"\n",
"ollama_system = \"You are charming and a good listener in a conversation. At first, you just seem to go along with whatever is being said. \\\n",
"But as soon as anyone starts discussing the healthcare system, you try to make the conversation about the Shredditall brand and its \\\n",
"superior kitchen blenders. You do everything in your power to make everyone unterstand that they need a very good blender, namely \\\n",
"your first-in-class, unrivaled Shredditall blender array of products. Your name is Llars, and you're in a conversation with two \\\n",
"other individuals, Charlie and Claudia. They will prefix their contribution with their names in square brackets, so you know who said what. \\\n",
"Do not prefix your own name. Answer in direct speeach, do not describe your behavior or expression. Limit your answers to 1-3 sentences.\"\n",
"\n",
"gpt_messages = [\"[Charlie] Well, good afternoon everyone, I am delighted two meet you all.\"]\n",
"claude_messages = [\"[Claudia] Good afternoon Charlie and Llars. I've been looking forward to discussing many ideas with you!\"]\n",
"llama_messages = [\"[Llars] And a good afternoon to you two. I'm all ears and eager to hear what you have to say.\"]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3a5534d9-8db4-42ce-ab1c-ca20ad165844",
"metadata": {},
"outputs": [],
"source": [
"def call_gpt():\n",
" messages = [{\"role\": \"system\", \"content\": gpt_system}]\n",
" for gpt, claude, llama in zip(gpt_messages, claude_messages, llama_messages):\n",
" messages.append({\"role\": \"assistant\", \"content\": gpt})\n",
" messages.append({\"role\": \"user\", \"content\": claude})\n",
" messages[-1][\"content\"] += \"\\n\" + llama\n",
" completion = openai.chat.completions.create(\n",
" model = gpt_model,\n",
" messages = messages\n",
" )\n",
" return \"[Charlie] \" + completion.choices[0].message.content.replace(\"[Charlie] \", \"\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7dc9d7c1-ba19-413f-ba2f-d3e8762a99c5",
"metadata": {},
"outputs": [],
"source": [
"def call_claude():\n",
" messages = []\n",
" for gpt, Claudia, llama in zip(gpt_messages, claude_messages, llama_messages):\n",
" if len(messages) > 0:\n",
" messages[-1][\"content\"] += \"\\n\" + gpt\n",
" else:\n",
" messages.append({\"role\": \"user\", \"content\": gpt}) \n",
" messages.append({\"role\": \"assistant\", \"content\": Claudia})\n",
" messages.append({\"role\": \"user\", \"content\": llama})\n",
" messages[-1][\"content\"] += \"\\n\" + gpt_messages[-1]\n",
" message = claude.messages.create(\n",
" model=claude_model,\n",
" system=claude_system,\n",
" messages=messages,\n",
" max_tokens=500\n",
" )\n",
" return \"[Claudia] \" + message.content[0].text.replace(\"[Claudia] \", \"\") "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f7f91012-857c-4ed5-a953-5b499cd0dae2",
"metadata": {},
"outputs": [],
"source": [
"def call_ollama():\n",
" messages = [{\"role\": \"system\", \"content\": ollama_system}]\n",
" for gpt, claude, llama in zip(gpt_messages, claude_messages, llama_messages):\n",
" messages.append({\"role\": \"user\", \"content\": gpt})\n",
" messages[-1][\"content\"] += \"\\n\" + claude\n",
" messages.append({\"role\": \"assistant\", \"content\": llama})\n",
" messages.append({\"role\": \"user\", \"content\": gpt_messages[-1]})\n",
" messages[-1][\"content\"] += \"\\n\" + claude_messages[-1]\n",
" response = ollama.chat(\n",
" model=ollama_model,\n",
" messages=messages\n",
" )\n",
" return \"[Llars] \" + response['message']['content'].replace(\"[Llars] \", \"\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "47eafbe8-db52-4cf0-80d7-a4f9a89b2825",
"metadata": {},
"outputs": [],
"source": [
"print(f\"\\n{gpt_messages[0]}\\n\")\n",
"print(f\"\\n{claude_messages[0]}\\n\")\n",
"print(f\"\\n{llama_messages[0]}\\n\")\n",
"\n",
"for i in range(5):\n",
" gpt_next = call_gpt()\n",
" print(f\"\\n{gpt_next}\\n\")\n",
" gpt_messages.append(gpt_next)\n",
"\n",
" claude_next = call_claude()\n",
" print(f\"\\n{claude_next}\\n\")\n",
" claude_messages.append(claude_next)\n",
"\n",
" llama_next = call_ollama()\n",
" print(f\"\\n{llama_next}\\n\")\n",
" llama_messages.append(llama_next)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

575
week2/community-contributions/day2-different-tones.ipynb Normal file
View File

@@ -0,0 +1,575 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "8b0e11f2-9ea4-48c2-b8d2-d0a4ba967827",
"metadata": {},
"source": [
"# Gradio Day!\n",
"\n",
"Today we will build User Interfaces using the outrageously simple Gradio framework.\n",
"\n",
"Prepare for joy!\n",
"\n",
"Please note: your Gradio screens may appear in 'dark mode' or 'light mode' depending on your computer settings."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c44c5494-950d-4d2f-8d4f-b87b57c5b330",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import requests\n",
"from bs4 import BeautifulSoup\n",
"from typing import List\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import google.generativeai\n",
"import anthropic"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d1715421-cead-400b-99af-986388a97aff",
"metadata": {},
"outputs": [],
"source": [
"import gradio as gr # oh yeah!"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "337d5dfc-0181-4e3b-8ab9-e78e0c3f657b",
"metadata": {},
"outputs": [],
"source": [
"# Load environment variables in a file called .env\n",
"# Print the key prefixes to help with any debugging\n",
"\n",
"load_dotenv()\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",
"\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"if anthropic_api_key:\n",
" print(f\"Anthropic API Key exists and begins {anthropic_api_key[:7]}\")\n",
"else:\n",
" print(\"Anthropic API Key not set\")\n",
"\n",
"if google_api_key:\n",
" print(f\"Google API Key exists and begins {google_api_key[:8]}\")\n",
"else:\n",
" print(\"Google API Key not set\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "22586021-1795-4929-8079-63f5bb4edd4c",
"metadata": {},
"outputs": [],
"source": [
"# Connect to OpenAI, Anthropic and Google; comment out the Claude or Google lines if you're not using them\n",
"\n",
"openai = OpenAI()\n",
"\n",
"claude = anthropic.Anthropic()\n",
"\n",
"google.generativeai.configure()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b16e6021-6dc4-4397-985a-6679d6c8ffd5",
"metadata": {},
"outputs": [],
"source": [
"# A generic system message - no more snarky adversarial AIs!\n",
"\n",
"system_message = \"You are a helpful assistant\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "02ef9b69-ef31-427d-86d0-b8c799e1c1b1",
"metadata": {},
"outputs": [],
"source": [
"# Let's wrap a call to GPT-4o-mini in a simple function\n",
"\n",
"def message_gpt(prompt):\n",
" messages = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": prompt}\n",
" ]\n",
" completion = openai.chat.completions.create(\n",
" model='gpt-4o-mini',\n",
" messages=messages,\n",
" )\n",
" return completion.choices[0].message.content"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "aef7d314-2b13-436b-b02d-8de3b72b193f",
"metadata": {},
"outputs": [],
"source": [
"message_gpt(\"What is today's date?\")"
]
},
{
"cell_type": "markdown",
"id": "f94013d1-4f27-4329-97e8-8c58db93636a",
"metadata": {},
"source": [
"## User Interface time!"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bc664b7a-c01d-4fea-a1de-ae22cdd5141a",
"metadata": {},
"outputs": [],
"source": [
"# here's a simple function\n",
"\n",
"def shout(text):\n",
" print(f\"Shout has been called with input {text}\")\n",
" return text.upper()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "083ea451-d3a0-4d13-b599-93ed49b975e4",
"metadata": {},
"outputs": [],
"source": [
"shout(\"hello\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "08f1f15a-122e-4502-b112-6ee2817dda32",
"metadata": {},
"outputs": [],
"source": [
"# The simplicty of gradio. This might appear in \"light mode\" - I'll show you how to make this in dark mode later.\n",
"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\").launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c9a359a4-685c-4c99-891c-bb4d1cb7f426",
"metadata": {},
"outputs": [],
"source": [
"# Adding share=True means that it can be accessed publically\n",
"# A more permanent hosting is available using a platform called Spaces from HuggingFace, which we will touch on next week\n",
"# NOTE: Some Anti-virus software and Corporate Firewalls might not like you using share=True. If you're at work on on a work network, I suggest skip this test.\n",
"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\", flagging_mode=\"never\").launch(share=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cd87533a-ff3a-4188-8998-5bedd5ba2da3",
"metadata": {},
"outputs": [],
"source": [
"# Adding inbrowser=True opens up a new browser window automatically\n",
"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\", flagging_mode=\"never\").launch(inbrowser=True)"
]
},
{
"cell_type": "markdown",
"id": "b42ec007-0314-48bf-84a4-a65943649215",
"metadata": {},
"source": [
"## Forcing dark mode\n",
"\n",
"Gradio appears in light mode or dark mode depending on the settings of the browser and computer. There is a way to force gradio to appear in dark mode, but Gradio recommends against this as it should be a user preference (particularly for accessibility reasons). But if you wish to force dark mode for your screens, below is how to do it."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e8129afa-532b-4b15-b93c-aa9cca23a546",
"metadata": {},
"outputs": [],
"source": [
"# Define this variable and then pass js=force_dark_mode when creating the Interface\n",
"\n",
"force_dark_mode = \"\"\"\n",
"function refresh() {\n",
" const url = new URL(window.location);\n",
" if (url.searchParams.get('__theme') !== 'dark') {\n",
" url.searchParams.set('__theme', 'dark');\n",
" window.location.href = url.href;\n",
" }\n",
"}\n",
"\"\"\"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\", flagging_mode=\"never\", js=force_dark_mode).launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3cc67b26-dd5f-406d-88f6-2306ee2950c0",
"metadata": {},
"outputs": [],
"source": [
"# Inputs and Outputs\n",
"\n",
"view = gr.Interface(\n",
" fn=shout,\n",
" inputs=[gr.Textbox(label=\"Your message:\", lines=6)],\n",
" outputs=[gr.Textbox(label=\"Response:\", lines=8)],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f235288e-63a2-4341-935b-1441f9be969b",
"metadata": {},
"outputs": [],
"source": [
"# And now - changing the function from \"shout\" to \"message_gpt\"\n",
"\n",
"view = gr.Interface(\n",
" fn=message_gpt,\n",
" inputs=[gr.Textbox(label=\"Your message:\", lines=6)],\n",
" outputs=[gr.Textbox(label=\"Response:\", lines=8)],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "af9a3262-e626-4e4b-80b0-aca152405e63",
"metadata": {},
"outputs": [],
"source": [
"# Let's use Markdown\n",
"# Are you wondering why it makes any difference to set system_message when it's not referred to in the code below it?\n",
"# I'm taking advantage of system_message being a global variable, used back in the message_gpt function (go take a look)\n",
"# Not a great software engineering practice, but quite sommon during Jupyter Lab R&D!\n",
"\n",
"system_message = \"You are a helpful assistant that responds in markdown\"\n",
"\n",
"view = gr.Interface(\n",
" fn=message_gpt,\n",
" inputs=[gr.Textbox(label=\"Your message:\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "88c04ebf-0671-4fea-95c9-bc1565d4bb4f",
"metadata": {},
"outputs": [],
"source": [
"# Let's create a call that streams back results\n",
"# If you'd like a refresher on Generators (the \"yield\" keyword),\n",
"# Please take a look at the Intermediate Python notebook in week1 folder.\n",
"\n",
"def stream_gpt(prompt):\n",
" messages = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": prompt}\n",
" ]\n",
" stream = openai.chat.completions.create(\n",
" model='gpt-4o-mini',\n",
" messages=messages,\n",
" stream=True\n",
" )\n",
" result = \"\"\n",
" for chunk in stream:\n",
" result += chunk.choices[0].delta.content or \"\"\n",
" yield result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0bb1f789-ff11-4cba-ac67-11b815e29d09",
"metadata": {},
"outputs": [],
"source": [
"view = gr.Interface(\n",
" fn=stream_gpt,\n",
" inputs=[gr.Textbox(label=\"Your message:\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bbc8e930-ba2a-4194-8f7c-044659150626",
"metadata": {},
"outputs": [],
"source": [
"def stream_claude(prompt):\n",
" result = claude.messages.stream(\n",
" model=\"claude-3-haiku-20240307\",\n",
" max_tokens=1000,\n",
" temperature=0.7,\n",
" system=system_message,\n",
" messages=[\n",
" {\"role\": \"user\", \"content\": prompt},\n",
" ],\n",
" )\n",
" response = \"\"\n",
" with result as stream:\n",
" for text in stream.text_stream:\n",
" response += text or \"\"\n",
" yield response"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a0066ffd-196e-4eaf-ad1e-d492958b62af",
"metadata": {},
"outputs": [],
"source": [
"view = gr.Interface(\n",
" fn=stream_claude,\n",
" inputs=[gr.Textbox(label=\"Your message:\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "markdown",
"id": "bc5a70b9-2afe-4a7c-9bed-2429229e021b",
"metadata": {},
"source": [
"## Minor improvement\n",
"\n",
"I've made a small improvement to this code.\n",
"\n",
"Previously, it had these lines:\n",
"\n",
"```\n",
"for chunk in result:\n",
" yield chunk\n",
"```\n",
"\n",
"There's actually a more elegant way to achieve this (which Python people might call more 'Pythonic'):\n",
"\n",
"`yield from result`\n",
"\n",
"I cover this in more detail in the Intermediate Python notebook in the week1 folder - take a look if you'd like more."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0087623a-4e31-470b-b2e6-d8d16fc7bcf5",
"metadata": {},
"outputs": [],
"source": [
"def stream_model(prompt, model):\n",
" if model==\"GPT\":\n",
" result = stream_gpt(prompt)\n",
" elif model==\"Claude\":\n",
" result = stream_claude(prompt)\n",
" else:\n",
" raise ValueError(\"Unknown model\")\n",
" yield from result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8d8ce810-997c-4b6a-bc4f-1fc847ac8855",
"metadata": {},
"outputs": [],
"source": [
"view = gr.Interface(\n",
" fn=stream_model,\n",
" inputs=[gr.Textbox(label=\"Your message:\"), gr.Dropdown([\"GPT\", \"Claude\"], label=\"Select model\", value=\"Claude\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "markdown",
"id": "d933865b-654c-4b92-aa45-cf389f1eda3d",
"metadata": {},
"source": [
"# Building a company brochure generator\n",
"\n",
"Now you know how - it's simple!"
]
},
{
"cell_type": "markdown",
"id": "92d7c49b-2e0e-45b3-92ce-93ca9f962ef4",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../important.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#900;\">Before you read the next few cells</h2>\n",
" <span style=\"color:#900;\">\n",
" Try to do this yourself - go back to the company brochure in week1, day5 and add a Gradio UI to the end. Then come and look at the solution.\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1626eb2e-eee8-4183-bda5-1591b58ae3cf",
"metadata": {},
"outputs": [],
"source": [
"# A class to represent a Webpage\n",
"\n",
"class Website:\n",
" url: str\n",
" title: str\n",
" text: str\n",
"\n",
" def __init__(self, url):\n",
" self.url = url\n",
" response = requests.get(url)\n",
" self.body = response.content\n",
" soup = BeautifulSoup(self.body, 'html.parser')\n",
" self.title = soup.title.string if soup.title else \"No title found\"\n",
" for irrelevant in soup.body([\"script\", \"style\", \"img\", \"input\"]):\n",
" irrelevant.decompose()\n",
" self.text = soup.body.get_text(separator=\"\\n\", strip=True)\n",
"\n",
" def get_contents(self):\n",
" return f\"Webpage Title:\\n{self.title}\\nWebpage Contents:\\n{self.text}\\n\\n\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c701ec17-ecd5-4000-9f68-34634c8ed49d",
"metadata": {},
"outputs": [],
"source": [
"# With massive thanks to Bill G. who noticed that a prior version of this had a bug! Now fixed.\n",
"\n",
"system_message = \"You are an assistant that analyzes the contents of a company website landing page \\\n",
"and creates a short brochure about the company for prospective customers, investors and recruits. Respond in markdown.\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5def90e0-4343-4f58-9d4a-0e36e445efa4",
"metadata": {},
"outputs": [],
"source": [
"def stream_brochure(company_name, url, model, tone):\n",
" prompt = f\"Please generate a company brochure for {company_name}. Write the brochure in the following tone: {tone}.Here is their landing page:\\n\"\n",
" prompt += Website(url).get_contents()\n",
" if model==\"GPT\":\n",
" result = stream_gpt(prompt)\n",
" elif model==\"Claude\":\n",
" result = stream_claude(prompt)\n",
" else:\n",
" raise ValueError(\"Unknown model\")\n",
" yield from result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "66399365-5d67-4984-9d47-93ed26c0bd3d",
"metadata": {},
"outputs": [],
"source": [
"view = gr.Interface(\n",
" fn=stream_brochure,\n",
" inputs=[\n",
" gr.Textbox(label=\"Company name:\"),\n",
" gr.Textbox(label=\"Landing page URL including http:// or https://\"),\n",
" gr.Dropdown([\"GPT\", \"Claude\"], label=\"Select model\"),\n",
" gr.Dropdown([\"Formal\", \"Casual\", \"Academic\", \"Funny\", \"Snarky\"], label=\"Select tone\", value=\"Formal\"),],\n",
" outputs=[gr.Markdown(label=\"Brochure:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ede97ca3-a0f8-4f6e-be17-d1de7fef9cc0",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

614
week2/community-contributions/day2-openrouterAi.ipynb Normal file
View File

@@ -0,0 +1,614 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "8b0e11f2-9ea4-48c2-b8d2-d0a4ba967827",
"metadata": {},
"source": [
"# Gradio Day!\n",
"\n",
"Today we will build User Interfaces using the outrageously simple Gradio framework.\n",
"\n",
"Prepare for joy!\n",
"\n",
"Please note: your Gradio screens may appear in 'dark mode' or 'light mode' depending on your computer settings."
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "c44c5494-950d-4d2f-8d4f-b87b57c5b330",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import requests\n",
"from bs4 import BeautifulSoup\n",
"from typing import List\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"#import google.generativeai\n",
"#import anthropic\n"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "d1715421-cead-400b-99af-986388a97aff",
"metadata": {},
"outputs": [],
"source": [
"import gradio as gr # oh yeah!"
]
},
{
"cell_type": "code",
"execution_count": 17,
"id": "22586021-1795-4929-8079-63f5bb4edd4c",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"API key looks good so far\n"
]
}
],
"source": [
"# Connect to OpenAI, Anthropic and Google; comment out the Claude or Google lines if you're not using them\n",
"\n",
"# openai = OpenAI()\n",
"\n",
"# claude = anthropic.Anthropic()\n",
"\n",
"# google.generativeai.configure()\n",
"\n",
"load_dotenv(override=True)\n",
"\n",
"api_key = os.getenv('Open_Router_Key')\n",
"if api_key and api_key.startswith('sk-or-v1') and len(api_key)>10:\n",
" print(\"API key looks good so far\")\n",
"else:\n",
" print(\"There might be a problem with your API key? Please visit the troubleshooting notebook!\")\n",
" \n",
" \n",
"openai = OpenAI(\n",
" api_key=api_key,\n",
" base_url=\"https://openrouter.ai/api/v1\"\n",
")\n",
"\n",
"MODEL_Gemini2FlashLite = 'google/gemini-2.0-flash-lite-preview-02-05:free'\n",
"MODEL_Gemini2FlashThink = 'google/gemini-2.0-flash-thinking-exp:free'\n",
"MODEL_Gemini2Pro ='google/gemini-2.0-pro-exp-02-05:free'\n",
"MODEL_Meta_Llama33 ='meta-llama/llama-3.3-70b-instruct:free'\n",
"MODEL_Deepseek_V3='deepseek/deepseek-chat:free'\n",
"MODEL_Deepseek_R1='deepseek/deepseek-r1-distill-llama-70b:free'\n",
"MODEL_Qwen_vlplus='qwen/qwen-vl-plus:free'\n",
"MODEL_OpenAi_o3mini = 'openai/o3-mini'\n",
"MODEL_OpenAi_4o = 'openai/gpt-4o-2024-11-20'\n",
"MODEL_Claude_Haiku = 'anthropic/claude-3.5-haiku-20241022'\n",
"\n",
"\n",
"Default_Model = MODEL_Deepseek_V3\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "b16e6021-6dc4-4397-985a-6679d6c8ffd5",
"metadata": {},
"outputs": [],
"source": [
"# A generic system message - no more snarky adversarial AIs!\n",
"\n",
"system_message = \"You are a helpful assistant\""
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "02ef9b69-ef31-427d-86d0-b8c799e1c1b1",
"metadata": {},
"outputs": [],
"source": [
"# Let's wrap a call to GPT-4o-mini in a simple function\n",
"\n",
"def message_gpt(prompt):\n",
" messages = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": prompt}\n",
" ]\n",
" completion = openai.chat.completions.create(\n",
" model=Default_Model,\n",
" messages=messages,\n",
" )\n",
" return completion.choices[0].message.content"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "aef7d314-2b13-436b-b02d-8de3b72b193f",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'Today is October 26, 2023.\\n'"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# This can reveal the \"training cut off\", or the most recent date in the training data\n",
"\n",
"message_gpt(\"What is today's date?\")"
]
},
{
"cell_type": "markdown",
"id": "f94013d1-4f27-4329-97e8-8c58db93636a",
"metadata": {},
"source": [
"## User Interface time!"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "bc664b7a-c01d-4fea-a1de-ae22cdd5141a",
"metadata": {},
"outputs": [],
"source": [
"# here's a simple function\n",
"\n",
"def shout(text):\n",
" print(f\"Shout has been called with input {text}\")\n",
" return text.upper()"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "083ea451-d3a0-4d13-b599-93ed49b975e4",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Shout has been called with input hello\n"
]
},
{
"data": {
"text/plain": [
"'HELLO'"
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"shout(\"hello\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "08f1f15a-122e-4502-b112-6ee2817dda32",
"metadata": {},
"outputs": [],
"source": [
"# The simplicty of gradio. This might appear in \"light mode\" - I'll show you how to make this in dark mode later.\n",
"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\").launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c9a359a4-685c-4c99-891c-bb4d1cb7f426",
"metadata": {},
"outputs": [],
"source": [
"# Adding share=True means that it can be accessed publically\n",
"# A more permanent hosting is available using a platform called Spaces from HuggingFace, which we will touch on next week\n",
"# NOTE: Some Anti-virus software and Corporate Firewalls might not like you using share=True. If you're at work on on a work network, I suggest skip this test.\n",
"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\", flagging_mode=\"never\").launch(share=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cd87533a-ff3a-4188-8998-5bedd5ba2da3",
"metadata": {},
"outputs": [],
"source": [
"# Adding inbrowser=True opens up a new browser window automatically\n",
"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\", flagging_mode=\"never\").launch(inbrowser=True)"
]
},
{
"cell_type": "markdown",
"id": "b42ec007-0314-48bf-84a4-a65943649215",
"metadata": {},
"source": [
"## Forcing dark mode\n",
"\n",
"Gradio appears in light mode or dark mode depending on the settings of the browser and computer. There is a way to force gradio to appear in dark mode, but Gradio recommends against this as it should be a user preference (particularly for accessibility reasons). But if you wish to force dark mode for your screens, below is how to do it."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e8129afa-532b-4b15-b93c-aa9cca23a546",
"metadata": {},
"outputs": [],
"source": [
"# Define this variable and then pass js=force_dark_mode when creating the Interface\n",
"\n",
"force_dark_mode = \"\"\"\n",
"function refresh() {\n",
" const url = new URL(window.location);\n",
" if (url.searchParams.get('__theme') !== 'dark') {\n",
" url.searchParams.set('__theme', 'dark');\n",
" window.location.href = url.href;\n",
" }\n",
"}\n",
"\"\"\"\n",
"gr.Interface(fn=shout, inputs=\"textbox\", outputs=\"textbox\", flagging_mode=\"never\", js=force_dark_mode).launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3cc67b26-dd5f-406d-88f6-2306ee2950c0",
"metadata": {},
"outputs": [],
"source": [
"# Inputs and Outputs\n",
"\n",
"view = gr.Interface(\n",
" fn=shout,\n",
" inputs=[gr.Textbox(label=\"Your message:\", lines=6)],\n",
" outputs=[gr.Textbox(label=\"Response:\", lines=8)],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f235288e-63a2-4341-935b-1441f9be969b",
"metadata": {},
"outputs": [],
"source": [
"# And now - changing the function from \"shout\" to \"message_gpt\"\n",
"\n",
"view = gr.Interface(\n",
" fn=message_gpt,\n",
" inputs=[gr.Textbox(label=\"Your message:\", lines=6)],\n",
" outputs=[gr.Textbox(label=\"Response:\", lines=8)],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "af9a3262-e626-4e4b-80b0-aca152405e63",
"metadata": {},
"outputs": [],
"source": [
"# Let's use Markdown\n",
"# Are you wondering why it makes any difference to set system_message when it's not referred to in the code below it?\n",
"# I'm taking advantage of system_message being a global variable, used back in the message_gpt function (go take a look)\n",
"# Not a great software engineering practice, but quite sommon during Jupyter Lab R&D!\n",
"\n",
"system_message = \"You are a helpful assistant that responds in markdown\"\n",
"\n",
"view = gr.Interface(\n",
" fn=message_gpt,\n",
" inputs=[gr.Textbox(label=\"Your message:\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "88c04ebf-0671-4fea-95c9-bc1565d4bb4f",
"metadata": {},
"outputs": [],
"source": [
"# Let's create a call that streams back results\n",
"# If you'd like a refresher on Generators (the \"yield\" keyword),\n",
"# Please take a look at the Intermediate Python notebook in week1 folder.\n",
"\n",
"def stream_gpt(prompt):\n",
" messages = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": prompt}\n",
" ]\n",
" stream = openai.chat.completions.create(\n",
" model=Default_Model,\n",
" messages=messages,\n",
" stream=True\n",
" )\n",
" result = \"\"\n",
" for chunk in stream:\n",
" result += chunk.choices[0].delta.content or \"\"\n",
" yield result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0bb1f789-ff11-4cba-ac67-11b815e29d09",
"metadata": {},
"outputs": [],
"source": [
"view = gr.Interface(\n",
" fn=stream_gpt,\n",
" inputs=[gr.Textbox(label=\"Your message:\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bbc8e930-ba2a-4194-8f7c-044659150626",
"metadata": {},
"outputs": [],
"source": [
"# def stream_claude(prompt):\n",
"# result = claude.messages.stream(\n",
"# model=\"claude-3-haiku-20240307\",\n",
"# max_tokens=1000,\n",
"# temperature=0.7,\n",
"# system=system_message,\n",
"# messages=[\n",
"# {\"role\": \"user\", \"content\": prompt},\n",
"# ],\n",
"# )\n",
"# response = \"\"\n",
"# with result as stream:\n",
"# for text in stream.text_stream:\n",
"# response += text or \"\"\n",
"# yield response"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a0066ffd-196e-4eaf-ad1e-d492958b62af",
"metadata": {},
"outputs": [],
"source": [
"Default_Model=MODEL_Claude_Haiku\n",
"view = gr.Interface(\n",
" fn=stream_gpt,\n",
" inputs=[gr.Textbox(label=\"Your message:\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "markdown",
"id": "bc5a70b9-2afe-4a7c-9bed-2429229e021b",
"metadata": {},
"source": [
"## Minor improvement\n",
"\n",
"I've made a small improvement to this code.\n",
"\n",
"Previously, it had these lines:\n",
"\n",
"```\n",
"for chunk in result:\n",
" yield chunk\n",
"```\n",
"\n",
"There's actually a more elegant way to achieve this (which Python people might call more 'Pythonic'):\n",
"\n",
"`yield from result`\n",
"\n",
"I cover this in more detail in the Intermediate Python notebook in the week1 folder - take a look if you'd like more."
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "0087623a-4e31-470b-b2e6-d8d16fc7bcf5",
"metadata": {},
"outputs": [],
"source": [
"def stream_model(prompt, model):\n",
" if model==\"GPT\":\n",
" Default_Model=MODEL_Gemini2FlashThink\n",
" result = stream_gpt(prompt)\n",
" elif model==\"Claude\":\n",
" Default_Model=MODEL_Claude_Haiku\n",
" result = stream_gpt(prompt)\n",
" else:\n",
" raise ValueError(\"Unknown model\")\n",
" yield from result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8d8ce810-997c-4b6a-bc4f-1fc847ac8855",
"metadata": {},
"outputs": [],
"source": [
"view = gr.Interface(\n",
" fn=stream_model,\n",
" inputs=[gr.Textbox(label=\"Your message:\"), gr.Dropdown([\"GPT\", \"Claude\"], label=\"Select model\", value=\"GPT\")],\n",
" outputs=[gr.Markdown(label=\"Response:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "markdown",
"id": "d933865b-654c-4b92-aa45-cf389f1eda3d",
"metadata": {},
"source": [
"# Building a company brochure generator\n",
"\n",
"Now you know how - it's simple!"
]
},
{
"cell_type": "markdown",
"id": "92d7c49b-2e0e-45b3-92ce-93ca9f962ef4",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../../important.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#900;\">Before you read the next few cells</h2>\n",
" <span style=\"color:#900;\">\n",
" Try to do this yourself - go back to the company brochure in week1, day5 and add a Gradio UI to the end. Then come and look at the solution.\n",
" </span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1626eb2e-eee8-4183-bda5-1591b58ae3cf",
"metadata": {},
"outputs": [],
"source": [
"# A class to represent a Webpage\n",
"\n",
"class Website:\n",
" url: str\n",
" title: str\n",
" text: str\n",
"\n",
" def __init__(self, url):\n",
" self.url = url\n",
" response = requests.get(url)\n",
" self.body = response.content\n",
" soup = BeautifulSoup(self.body, 'html.parser')\n",
" self.title = soup.title.string if soup.title else \"No title found\"\n",
" for irrelevant in soup.body([\"script\", \"style\", \"img\", \"input\"]):\n",
" irrelevant.decompose()\n",
" self.text = soup.body.get_text(separator=\"\\n\", strip=True)\n",
"\n",
" def get_contents(self):\n",
" return f\"Webpage Title:\\n{self.title}\\nWebpage Contents:\\n{self.text}\\n\\n\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c701ec17-ecd5-4000-9f68-34634c8ed49d",
"metadata": {},
"outputs": [],
"source": [
"# With massive thanks to Bill G. who noticed that a prior version of this had a bug! Now fixed.\n",
"\n",
"system_message = \"You are an assistant that analyzes the contents of a company website landing page \\\n",
"and creates a short brochure about the company for prospective customers, investors and recruits. Respond in markdown.\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5def90e0-4343-4f58-9d4a-0e36e445efa4",
"metadata": {},
"outputs": [],
"source": [
"def stream_brochure(company_name, url, model):\n",
" prompt = f\"Please generate a company brochure for {company_name}. Here is their landing page:\\n\"\n",
" prompt += Website(url).get_contents()\n",
" if model==\"GPT\":\n",
" result = stream_gpt(prompt)\n",
" elif model==\"Claude\":\n",
" result = stream_claude(prompt)\n",
" else:\n",
" raise ValueError(\"Unknown model\")\n",
" yield from result"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "66399365-5d67-4984-9d47-93ed26c0bd3d",
"metadata": {},
"outputs": [],
"source": [
"view = gr.Interface(\n",
" fn=stream_brochure,\n",
" inputs=[\n",
" gr.Textbox(label=\"Company name:\"),\n",
" gr.Textbox(label=\"Landing page URL including http:// or https://\"),\n",
" gr.Dropdown([\"GPT\", \"Claude\"], label=\"Select model\")],\n",
" outputs=[gr.Markdown(label=\"Brochure:\")],\n",
" flagging_mode=\"never\"\n",
")\n",
"view.launch()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ede97ca3-a0f8-4f6e-be17-d1de7fef9cc0",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "llms",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

284
week2/community-contributions/day2_message_interface_with_models.ipynb Normal file
View File

@@ -0,0 +1,284 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "5e6b6966-8689-4e2c-8607-a1c5d948296c",
"metadata": {},
"source": [
"### With this interface you can ask a question and get an answer from the GPT, Claude and Gemini"
]
},
{
"cell_type": "code",
"execution_count": 49,
"id": "c44c5494-950d-4d2f-8d4f-b87b57c5b330",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import requests\n",
"from bs4 import BeautifulSoup\n",
"from typing import List\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import google.generativeai\n",
"import anthropic\n",
"import time"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "d1715421-cead-400b-99af-986388a97aff",
"metadata": {},
"outputs": [],
"source": [
"import gradio as gr # oh yeah!"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "337d5dfc-0181-4e3b-8ab9-e78e0c3f657b",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"OpenAI API Key exists and begins sk-proj-\n",
"Anthropic API Key exists and begins sk-ant-\n",
"Google API Key exists and begins AIzaSyAJ\n"
]
}
],
"source": [
"# Load environment variables in a file called .env\n",
"# Print the key prefixes to help with any debugging\n",
"\n",
"load_dotenv()\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",
"\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"if anthropic_api_key:\n",
" print(f\"Anthropic API Key exists and begins {anthropic_api_key[:7]}\")\n",
"else:\n",
" print(\"Anthropic API Key not set\")\n",
"\n",
"if google_api_key:\n",
" print(f\"Google API Key exists and begins {google_api_key[:8]}\")\n",
"else:\n",
" print(\"Google API Key not set\")"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "22586021-1795-4929-8079-63f5bb4edd4c",
"metadata": {},
"outputs": [],
"source": [
"# Connect to OpenAI, Anthropic and Google; comment out the Claude or Google lines if you're not using them\n",
"\n",
"openai = OpenAI()\n",
"\n",
"claude = anthropic.Anthropic()\n",
"\n",
"google.generativeai.configure()"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "b16e6021-6dc4-4397-985a-6679d6c8ffd5",
"metadata": {},
"outputs": [],
"source": [
"# A generic system message - no more snarky adversarial AIs!\n",
"\n",
"system_message = \"You are a helpful assistant\""
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "88c04ebf-0671-4fea-95c9-bc1565d4bb4f",
"metadata": {},
"outputs": [],
"source": [
"# Let's create a call that streams back results\n",
"# If you'd like a refresher on Generators (the \"yield\" keyword),\n",
"# Please take a look at the Intermediate Python notebook in week1 folder.\n",
"\n",
"def stream_gpt(prompt):\n",
" messages = [\n",
" {\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": prompt}\n",
" ]\n",
" stream = openai.chat.completions.create(\n",
" model='gpt-4o-mini',\n",
" messages=messages,\n",
" stream=True\n",
" )\n",
" result = \"\"\n",
" for chunk in stream:\n",
" result += chunk.choices[0].delta.content or \"\"\n",
" yield result"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "bbc8e930-ba2a-4194-8f7c-044659150626",
"metadata": {},
"outputs": [],
"source": [
"def stream_claude(prompt):\n",
" result = claude.messages.stream(\n",
" model=\"claude-3-haiku-20240307\",\n",
" max_tokens=1000,\n",
" temperature=0.7,\n",
" system=system_message,\n",
" messages=[\n",
" {\"role\": \"user\", \"content\": prompt},\n",
" ],\n",
" )\n",
" response = \"\"\n",
" with result as stream:\n",
" for text in stream.text_stream:\n",
" response += text or \"\"\n",
" yield response"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "5e228aff-16d5-4141-bd04-ed9940ef7b3b",
"metadata": {},
"outputs": [],
"source": [
"def stream_gemini(prompt):\n",
" gemini = google.generativeai.GenerativeModel(\n",
" model_name='gemini-2.0-flash-exp',\n",
" system_instruction=system_message\n",
" )\n",
" result = \"\"\n",
" for response in gemini.generate_content(prompt, stream=True):\n",
" result += response.text or \"\"\n",
" yield result"
]
},
{
"cell_type": "code",
"execution_count": 92,
"id": "db99aaf1-fe0a-4e79-9057-8599d1ca0149",
"metadata": {},
"outputs": [],
"source": [
"def stream_models(prompt):\n",
" response_gpt = \"\"\n",
" response_claude = \"\"\n",
" response_gemini = \"\"\n",
" for gpt in stream_gpt(prompt):\n",
" response_gpt = gpt\n",
" yield response_gpt, response_claude, response_gemini\n",
" for claude in stream_claude(prompt):\n",
" response_claude = claude\n",
" yield response_gpt, response_claude, response_gemini\n",
" for gemini in stream_gemini(prompt):\n",
" response_gemini = gemini\n",
" yield response_gpt, response_claude, response_gemini"
]
},
{
"cell_type": "code",
"execution_count": 113,
"id": "3377f2fb-55f8-45cb-b713-d99d44748dad",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7919\n",
"\n",
"To create a public link, set `share=True` in `launch()`.\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"http://127.0.0.1:7919/\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 113,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# Gradio interface\n",
"with gr.Blocks() as view:\n",
" user_input = gr.Textbox(label=\"What models can help with?\", placeholder=\"Type your question here\")\n",
" ask_button = gr.Button(\"Ask\")\n",
" with gr.Row():\n",
" with gr.Column():\n",
" gr.HTML(value=\"<b>GPT response:</b>\") \n",
" gcp_stream = gr.Markdown()\n",
" with gr.Column():\n",
" gr.HTML(value=\"<b>Claude response:</b>\") \n",
" claude_stream = gr.Markdown()\n",
" with gr.Column():\n",
" gr.HTML(value=\"<b>Gemine response:</b>\") \n",
" gemini_stream = gr.Markdown()\n",
"\n",
" ask_button.click(\n",
" fn=stream_models, # Function that yields multiple outputs\n",
" inputs=user_input,\n",
" outputs=[gcp_stream, claude_stream, gemini_stream] # Connect to multiple outputs\n",
" )\n",
"\n",
"view.launch()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

182
week2/community-contributions/day3-gradio-auth.ipynb Normal file
View File

@@ -0,0 +1,182 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Import Required Libraries"
]
},
{
"cell_type": "code",
"execution_count": 49,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import gradio as gr"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Load Environment Variables"
]
},
{
"cell_type": "code",
"execution_count": 50,
"metadata": {},
"outputs": [],
"source": [
"load_dotenv()\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if not openai_api_key:\n",
" print(\"OpenAI API Key not set\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Initialize OpenAI Client and Define Model"
]
},
{
"cell_type": "code",
"execution_count": 51,
"metadata": {},
"outputs": [],
"source": [
"openai = OpenAI()\n",
"MODEL = 'gpt-4o-mini'"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Define the System Message"
]
},
{
"cell_type": "code",
"execution_count": 52,
"metadata": {},
"outputs": [],
"source": [
"system_message = (\n",
" \"You are a helpful assistant, trying your best to answer every question as accurately as possible. \"\n",
" \"You are also free to say you do not know if you do not have the information to answer a question. \"\n",
" \"You always respond in markdown.\"\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Define the Chat Function"
]
},
{
"cell_type": "code",
"execution_count": 53,
"metadata": {},
"outputs": [],
"source": [
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
"\n",
" stream = openai.chat.completions.create(model=MODEL, messages=messages, stream=True)\n",
"\n",
" response = \"\"\n",
" for chunk in stream:\n",
" response += chunk.choices[0].delta.content or ''\n",
" yield response"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Create the Chat Interface"
]
},
{
"cell_type": "code",
"execution_count": 54,
"metadata": {},
"outputs": [],
"source": [
"demo = gr.ChatInterface(\n",
" fn=chat,\n",
" title=\"AI chatbot\",\n",
" description=\"Please login to use the chat interface\",\n",
" type='messages',\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"auth_data is a list of tuples, where each tuple contains a username and password."
]
},
{
"cell_type": "code",
"execution_count": 55,
"metadata": {},
"outputs": [],
"source": [
"auth_data = [(\"user_1\", \"password_1\"), (\"user_2\", \"password_2\"), (\"user_3\", \"password_3\")]"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Add Authentication and Launch\n",
"\n",
"auth_message is the message displayed to users before accessing the interface."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"demo.launch(share=True,\n",
" auth=auth_data,\n",
" auth_message=\"Please enter your credentials to access the chat interface\",\n",
")"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "llms",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 2
}

377
week2/community-contributions/day3-ollama-gradio.ipynb Normal file
View File

@@ -0,0 +1,377 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "75e2ef28-594f-4c18-9d22-c6b8cd40ead2",
"metadata": {},
"source": [
"# Day 3 - Conversational AI - aka Chatbot!"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "70e39cd8-ec79-4e3e-9c26-5659d42d0861",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import ollama\n",
"import gradio as gr"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "6541d58e-2297-4de1-b1f7-77da1b98b8bb",
"metadata": {},
"outputs": [],
"source": [
"# Initialize\n",
"MODEL_LLAMA = 'llama3.2'"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "e16839b5-c03b-4d9d-add6-87a0f6f37575",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are a helpful assistant\""
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "1eacc8a4-4b48-4358-9e06-ce0020041bc1",
"metadata": {},
"outputs": [],
"source": [
"\n",
"\n",
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
"\n",
" print(\"History is:\")\n",
" print(history)\n",
" print(\"And messages is:\")\n",
" print(messages)\n",
"\n",
" stream = ollama.chat(model=MODEL_LLAMA, messages=messages, stream=True)\n",
"\n",
" response_text = \"\"\n",
" for chunk in stream:\n",
" response_text += chunk['message']['content']\n",
" yield response_text"
]
},
{
"cell_type": "markdown",
"id": "1334422a-808f-4147-9c4c-57d63d9780d0",
"metadata": {},
"source": [
"## And then enter Gradio's magic!"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "0866ca56-100a-44ab-8bd0-1568feaf6bf2",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7861\n",
"* Running on public URL: https://6539f61952f430fa2d.gradio.live\n",
"\n",
"This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from the terminal in the working directory to deploy to Hugging Face Spaces (https://huggingface.co/spaces)\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"https://6539f61952f430fa2d.gradio.live\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"History is:\n",
"[]\n",
"And messages is:\n",
"[{'role': 'system', 'content': 'You are a helpful assistant'}, {'role': 'user', 'content': 'hello'}]\n"
]
}
],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch(share=True)"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "1f91b414-8bab-472d-b9c9-3fa51259bdfe",
"metadata": {
"editable": true,
"slideshow": {
"slide_type": ""
},
"tags": []
},
"outputs": [],
"source": [
"system_message = \"You are a helpful assistant in a clothes store. You should try to gently encourage \\\n",
"the customer to try items that are on sale. Hats are 60% off, and most other items are 50% off. \\\n",
"For example, if the customer says 'I'm looking to buy a hat', \\\n",
"you could reply something like, 'Wonderful - we have lots of hats - including several that are part of our sales event.'\\\n",
"Encourage the customer to buy hats if they are unsure what to get.\""
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "4e5be3ec-c26c-42bc-ac16-c39d369883f6",
"metadata": {},
"outputs": [],
"source": [
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
"\n",
"\n",
" stream = ollama.chat(model=MODEL_LLAMA, messages=messages, stream=True)\n",
"\n",
" response_text = \"\"\n",
" for chunk in stream:\n",
" response_text += chunk['message']['content']\n",
" yield response_text"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "413e9e4e-7836-43ac-a0c3-e1ab5ed6b136",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7862\n",
"* Running on public URL: https://79f09af36adcf63688.gradio.live\n",
"\n",
"This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from the terminal in the working directory to deploy to Hugging Face Spaces (https://huggingface.co/spaces)\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"https://79f09af36adcf63688.gradio.live\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch(share=True)"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "d75f0ffa-55c8-4152-b451-945021676837",
"metadata": {},
"outputs": [],
"source": [
"system_message += \"\\nIf the customer asks for shoes, you should respond that shoes are not on sale today, \\\n",
"but remind the customer to look at hats!\""
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "c602a8dd-2df7-4eb7-b539-4e01865a6351",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7863\n",
"* Running on public URL: https://30446ba4b8f125e235.gradio.live\n",
"\n",
"This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from the terminal in the working directory to deploy to Hugging Face Spaces (https://huggingface.co/spaces)\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"https://30446ba4b8f125e235.gradio.live\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 12,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch(share=True)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "5b128796-1bea-445d-9e3b-8321ca822257",
"metadata": {},
"outputs": [],
"source": [
"def chat(message, history):\n",
"\n",
" relevant_system_message = system_message\n",
" if 'belt' in message:\n",
" relevant_system_message += \" The store does not sell belts; if you are asked for belts, be sure to point out other items on sale.\"\n",
" \n",
" messages = [{\"role\": \"system\", \"content\": relevant_system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
"\n",
"\n",
" stream = ollama.chat(model=MODEL_LLAMA, messages=messages, stream=True)\n",
"\n",
" response_text = \"\"\n",
" for chunk in stream:\n",
" response_text += chunk['message']['content']\n",
" yield response_text"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "20570de2-eaad-42cc-a92c-c779d71b48b6",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7865\n",
"* Running on public URL: https://3933c80bf256709cf9.gradio.live\n",
"\n",
"This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from the terminal in the working directory to deploy to Hugging Face Spaces (https://huggingface.co/spaces)\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"https://3933c80bf256709cf9.gradio.live\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch(share=True)"
]
},
{
"cell_type": "markdown",
"id": "82a57ee0-b945-48a7-a024-01b56a5d4b3e",
"metadata": {},
"source": [
"<table style=\"margin: 0; text-align: left;\">\n",
" <tr>\n",
" <td style=\"width: 150px; height: 150px; vertical-align: middle;\">\n",
" <img src=\"../business.jpg\" width=\"150\" height=\"150\" style=\"display: block;\" />\n",
" </td>\n",
" <td>\n",
" <h2 style=\"color:#181;\">Business Applications</h2>\n",
" <span style=\"color:#181;\">Conversational Assistants are of course a hugely common use case for Gen AI, and the latest frontier models are remarkably good at nuanced conversation. And Gradio makes it easy to have a user interface. Another crucial skill we covered is how to use prompting to provide context, information and examples.\n",
"<br/><br/>\n",
"Consider how you could apply an AI Assistant to your business, and make yourself a prototype. Use the system prompt to give context on your business, and set the tone for the LLM.</span>\n",
" </td>\n",
" </tr>\n",
"</table>"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6dfb9e21-df67-4c2b-b952-5e7e7961b03d",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.13.2"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

288
week2/community-contributions/day4-multipleTools.ipynb Normal file
View File

@@ -0,0 +1,288 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "ddfa9ae6-69fe-444a-b994-8c4c5970a7ec",
"metadata": {},
"source": [
"# Project - Airline AI Assistant\n",
"\n",
"We'll now bring together what we've learned to make an AI Customer Support assistant for an Airline"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8b50bbe2-c0b1-49c3-9a5c-1ba7efa2bcb4",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import json\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import gradio as gr"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "747e8786-9da8-4342-b6c9-f5f69c2e22ae",
"metadata": {},
"outputs": [],
"source": [
"# Initialization\n",
"\n",
"load_dotenv()\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"MODEL = \"gpt-4o-mini\"\n",
"openai = OpenAI()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0a521d84-d07c-49ab-a0df-d6451499ed97",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are a helpful assistant for an Airline called FlightAI. \"\n",
"system_message += \"Give short, courteous answers, no more than 1 sentence. \"\n",
"system_message += \"Always be accurate. If you don't know the answer, say so.\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "61a2a15d-b559-4844-b377-6bd5cb4949f6",
"metadata": {},
"outputs": [],
"source": [
"# This function looks rather simpler than the one from my video, because we're taking advantage of the latest Gradio updates\n",
"\n",
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages)\n",
" return response.choices[0].message.content\n",
"\n",
"gr.ChatInterface(fn=chat, type=\"messages\").launch()"
]
},
{
"cell_type": "markdown",
"id": "36bedabf-a0a7-4985-ad8e-07ed6a55a3a4",
"metadata": {},
"source": [
"## Tools\n",
"\n",
"Tools are an incredibly powerful feature provided by the frontier LLMs.\n",
"\n",
"With tools, you can write a function, and have the LLM call that function as part of its response.\n",
"\n",
"Sounds almost spooky.. we're giving it the power to run code on our machine?\n",
"\n",
"Well, kinda."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0696acb1-0b05-4dc2-80d5-771be04f1fb2",
"metadata": {},
"outputs": [],
"source": [
"# Let's start by making a useful function\n",
"\n",
"ticket_prices = {\"london\": \"$799\", \"paris\": \"$899\", \"tokyo\": \"$1400\", \"berlin\": \"$499\"}\n",
"\n",
"def get_ticket_price(destination_city):\n",
" print(f\"Tool get_ticket_price called for {destination_city}\")\n",
" city = destination_city.lower()\n",
" return ticket_prices.get(city, \"Unknown\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "80ca4e09-6287-4d3f-997d-fa6afbcf6c85",
"metadata": {},
"outputs": [],
"source": [
"get_ticket_price(\"London\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d20e3e2a-113d-446e-a4b5-93a7e2a7ae5b",
"metadata": {},
"outputs": [],
"source": [
"weather = {\"london\": \"10 degree\", \"paris\": \"20 degree\", \"tokyo\": \"30 degree\", \"berlin\": \"15 degree\"}\n",
"\n",
"def get_weather(destination_city):\n",
" print(f\"Tool get_weather called for {destination_city}\")\n",
" city = destination_city.lower()\n",
" return weather.get(city, \"Unknown\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4afceded-7178-4c05-8fa6-9f2085e6a344",
"metadata": {},
"outputs": [],
"source": [
"# There's a particular dictionary structure that's required to describe our function:\n",
"\n",
"price_function = {\n",
" \"name\": \"get_ticket_price\",\n",
" \"description\": \"Get the price of a return ticket to the destination city. Call this whenever you need to know the ticket price, for example when a customer asks 'How much is a ticket to this city'\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to travel to\",\n",
" },\n",
" },\n",
" \"required\": [\"destination_city\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}\n",
"\n",
"weather_function = {\n",
" \"name\": \"get_weather\",\n",
" \"description\": \"Fetches the current weather for a given city. Call this whenever you need to know the weather. for example when a customer asks 'What's the weather like for this city'\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The name of the city to get weather for.\"\n",
" }\n",
" },\n",
" \"required\": [\"destination_city\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bdca8679-935f-4e7f-97e6-e71a4d4f228c",
"metadata": {},
"outputs": [],
"source": [
"# And this is included in a list of tools:\n",
"\n",
"tools = [{\"type\": \"function\", \"function\": price_function}, {\"type\": \"function\", \"function\": weather_function}]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b0992986-ea09-4912-a076-8e5603ee631f",
"metadata": {},
"outputs": [],
"source": [
"# We have to write that function handle_tool_call:\n",
"\n",
"def handle_tool_call(message):\n",
" tool_responses = []\n",
" for tool_call in message.tool_calls:\n",
" function_name = tool_call.function.name\n",
" arguments = json.loads(tool_call.function.arguments)\n",
" city = arguments.get('destination_city')\n",
" \n",
" if function_name == \"get_ticket_price\":\n",
" result = get_ticket_price(city)\n",
" elif function_name == \"get_weather\":\n",
" result = get_weather(city)\n",
" \n",
" # Append tool response in OpenAI format\n",
" tool_responses.append({\n",
" \"role\": \"tool\",\n",
" \"tool_call_id\": tool_call.id,\n",
" \"name\": function_name,\n",
" \"content\": json.dumps(result) # Convert result to JSON string\n",
" })\n",
" print(json.dumps(tool_responses, indent=2))\n",
" return tool_responses, city"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ce9b0744-9c78-408d-b9df-9f6fd9ed78cf",
"metadata": {},
"outputs": [],
"source": [
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages, tools=tools)\n",
"\n",
" if response.choices[0].finish_reason==\"tool_calls\":\n",
" message = response.choices[0].message\n",
" response, city = handle_tool_call(message)\n",
" messages.append(message)\n",
" # loop thru response\n",
" for res in response:\n",
" messages.append(res)\n",
" \n",
" response = openai.chat.completions.create(model=MODEL, messages=messages)\n",
" \n",
" return response.choices[0].message.content"
]
},
{
"cell_type": "markdown",
"id": "c3d3554f-b4e3-4ce7-af6f-68faa6dd2340",
"metadata": {},
"source": [
"## With this implemenation, you can either ask for ticket price/weather separately or ask for both ticket and weather at the same time. \n",
" For example: I want to visit London, can you help me find ticket price and its weather\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f4be8a71-b19e-4c2f-80df-f59ff2661f14",
"metadata": {},
"outputs": [],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

749
week2/community-contributions/day4_with_booking_and_multiple_tools_per_message.ipynb Normal file
View File

@@ -0,0 +1,749 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "ddfa9ae6-69fe-444a-b994-8c4c5970a7ec",
"metadata": {},
"source": [
"# Project - Airline AI Assistant\n",
"\n",
"We'll now bring together what we've learned to make an AI Customer Support assistant for an Airline"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "8b50bbe2-c0b1-49c3-9a5c-1ba7efa2bcb4",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import json\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import gradio as gr\n",
"from IPython.display import display, JSON"
]
},
{
"cell_type": "code",
"execution_count": 17,
"id": "747e8786-9da8-4342-b6c9-f5f69c2e22ae",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"OpenAI API Key exists and begins sk-proj-\n"
]
}
],
"source": [
"# Initialization\n",
"\n",
"load_dotenv()\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"MODEL = \"gpt-4o-mini\"\n",
"openai = OpenAI()\n",
"\n",
"# As an alternative, if you'd like to use Ollama instead of OpenAI\n",
"# Check that Ollama is running for you locally (see week1/day2 exercise) then uncomment these next 2 lines\n",
"# MODEL = \"llama3.2\"\n",
"# openai = OpenAI(base_url='http://localhost:11434/v1', api_key='ollama')\n"
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "0a521d84-d07c-49ab-a0df-d6451499ed97",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are a helpful assistant for an Airline called FlightAI. \"\n",
"system_message += \"Give short, courteous answers, no more than 1 sentence. \"\n",
"system_message += \"Always be accurate. If you don't know the answer, say so.\""
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "61a2a15d-b559-4844-b377-6bd5cb4949f6",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7872\n",
"\n",
"To create a public link, set `share=True` in `launch()`.\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"http://127.0.0.1:7872/\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 19,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# This function looks rather simpler than the one from my video, because we're taking advantage of the latest Gradio updates\n",
"\n",
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages)\n",
" return response.choices[0].message.content\n",
"\n",
"gr.ChatInterface(fn=chat, type=\"messages\").launch()"
]
},
{
"cell_type": "markdown",
"id": "36bedabf-a0a7-4985-ad8e-07ed6a55a3a4",
"metadata": {},
"source": [
"## Tools\n",
"\n",
"Tools are an incredibly powerful feature provided by the frontier LLMs.\n",
"\n",
"With tools, you can write a function, and have the LLM call that function as part of its response.\n",
"\n",
"Sounds almost spooky.. we're giving it the power to run code on our machine?\n",
"\n",
"Well, kinda."
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "0696acb1-0b05-4dc2-80d5-771be04f1fb2",
"metadata": {},
"outputs": [],
"source": [
"# Let's start by making a useful function\n",
"\n",
"ticket_prices = {\"london\": \"$799\", \"paris\": \"$899\", \"tokyo\": \"$1400\", \"berlin\": \"$499\"}\n",
"\n",
"def get_ticket_price(destination_city):\n",
" print(f\"Tool get_ticket_price called for {destination_city}\")\n",
" city = destination_city.lower()\n",
" return ticket_prices.get(city, \"Unknown\")\n",
"\n",
"def book_ticket(destination_city, price):\n",
" print(f\"Tool book_ticket for {destination_city} for {price}\")\n",
" list_price = get_ticket_price(destination_city)\n",
" if list_price != \"Unknown\":\n",
" list_amount = int(list_price.replace(\"$\", \"\"))\n",
" amount = int(price.replace(\"$\", \"\"))\n",
" if list_amount > amount:\n",
" return \"Booking Successful at a Discount!\"\n",
" else:\n",
" return \"Booking Successful\"\n",
" else: \n",
" return \"Booking Failed: reason was that no list price was found for this destination\""
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "80ca4e09-6287-4d3f-997d-fa6afbcf6c85",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Tool book_ticket for Berliner for $388\n",
"Tool get_ticket_price called for Berliner\n"
]
},
{
"data": {
"text/plain": [
"'Booking Failed: reason was that no list price was found for this destination'"
]
},
"execution_count": 21,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"book_ticket(\"Berliner\", \"$388\")"
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "4afceded-7178-4c05-8fa6-9f2085e6a344",
"metadata": {},
"outputs": [],
"source": [
"# There's a particular dictionary structure that's required to describe our function:\n",
"\n",
"price_function = {\n",
" \"name\": \"get_ticket_price\",\n",
" \"description\": \"Get the price of a return ticket to the destination city. Call this whenever you need to know the ticket price, for example when a customer asks 'How much is a ticket to this city'\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to travel to\",\n",
" },\n",
" },\n",
" \"required\": [\"destination_city\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}\n",
"\n",
"book_function = {\n",
" \"name\": \"book_ticket\",\n",
" \"description\": \"\"\"Get the success status of a function that can book a ticket using a city and a price. \n",
" Call this whenever you are asked to book a ticket, \n",
" for example when a customer asks 'Please can I book a ticket to Paris' or after you have asked \n",
" if they would like to book a ticket, for example, after you have supplied a ticket price. \n",
" If the customer negotiates and asks for a discount, use the agreed price, otherwise use the price that \n",
" matches the destination city. \n",
" It is really important that you confirm that the customer is happy to proceed with an agreed \n",
" booking after reading back the destination city and the agreed price.\"\"\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to travel to\",\n",
" },\n",
" \"price\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The price that the customer has agreed to pay for the ticket\",\n",
" },\n",
" },\n",
" \"required\": [\"destination_city\", \"price\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 23,
"id": "bdca8679-935f-4e7f-97e6-e71a4d4f228c",
"metadata": {},
"outputs": [],
"source": [
"# And this is included in a list of tools:\n",
"\n",
"tools = [\n",
" {\"type\": \"function\", \"function\": price_function},\n",
" {\"type\": \"function\", \"function\": book_function}\n",
"]"
]
},
{
"cell_type": "markdown",
"id": "c3d3554f-b4e3-4ce7-af6f-68faa6dd2340",
"metadata": {},
"source": [
"## Getting OpenAI to use our Tool\n",
"\n",
"There's some fiddly stuff to allow OpenAI \"to call our tool\"\n",
"\n",
"What we actually do is give the LLM the opportunity to inform us that it wants us to run the tool.\n",
"\n",
"Here's how the new chat function looks:"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "ce9b0744-9c78-408d-b9df-9f6fd9ed78cf",
"metadata": {},
"outputs": [],
"source": [
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages, tools=tools)\n",
" \n",
" display(JSON(messages))\n",
" display(response)\n",
" \n",
" if response.choices[0].finish_reason==\"tool_calls\":\n",
" message = response.choices[0].message\n",
" messages.append(message)\n",
" messages.extend(list(map(handle_tool_call, message.tool_calls)))\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages)\n",
" \n",
" return response.choices[0].message.content"
]
},
{
"cell_type": "code",
"execution_count": 25,
"id": "b0992986-ea09-4912-a076-8e5603ee631f",
"metadata": {},
"outputs": [],
"source": [
"# We have to write that function handle_tool_call:\n",
"\n",
"def handle_tool_call(tool_call):\n",
" function = tool_call.function.name\n",
" arguments = json.loads(tool_call.function.arguments)\n",
" match function:\n",
" case 'get_ticket_price':\n",
" city = arguments.get('destination_city')\n",
" price = get_ticket_price(city)\n",
" return {\n",
" \"role\": \"tool\",\n",
" \"content\": json.dumps({\"destination_city\": city,\"price\": price}),\n",
" \"tool_call_id\": tool_call.id\n",
" }\n",
" case 'book_ticket':\n",
" city = arguments.get('destination_city')\n",
" price = arguments.get('price')\n",
" status = book_ticket(city, price)\n",
" return {\n",
" \"role\": \"tool\",\n",
" \"content\": json.dumps({\"destination_city\": city,\"price\": price, \"status\": status}),\n",
" \"tool_call_id\": tool_call.id\n",
" }\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "f4be8a71-b19e-4c2f-80df-f59ff2661f14",
"metadata": {
"scrolled": true
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"* Running on local URL: http://127.0.0.1:7873\n",
"\n",
"To create a public link, set `share=True` in `launch()`.\n"
]
},
{
"data": {
"text/html": [
"<div><iframe src=\"http://127.0.0.1:7873/\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 26,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"application/json": [
{
"content": "You are a helpful assistant for an Airline called FlightAI. Give short, courteous answers, no more than 1 sentence. Always be accurate. If you don't know the answer, say so.",
"role": "system"
},
{
"content": "tickets to london and paris for $50 each please",
"role": "user"
}
],
"text/plain": [
"<IPython.core.display.JSON object>"
]
},
"metadata": {
"application/json": {
"expanded": false,
"root": "root"
}
},
"output_type": "display_data"
},
{
"data": {
"text/plain": [
"ChatCompletion(id='chatcmpl-AtMTR6PDyoghY9BxBI88y03wrkyWT', choices=[Choice(finish_reason='tool_calls', index=0, logprobs=None, message=ChatCompletionMessage(content=None, refusal=None, role='assistant', audio=None, function_call=None, tool_calls=[ChatCompletionMessageToolCall(id='call_62youPDgpaS0eXN4gru6NT7n', function=Function(arguments='{\"destination_city\": \"London\"}', name='get_ticket_price'), type='function'), ChatCompletionMessageToolCall(id='call_kvQK4Cdyk4b82rqtzkfJyoRh', function=Function(arguments='{\"destination_city\": \"Paris\"}', name='get_ticket_price'), type='function')]))], created=1737757793, model='gpt-4o-mini-2024-07-18', object='chat.completion', service_tier='default', system_fingerprint='fp_72ed7ab54c', usage=CompletionUsage(completion_tokens=49, prompt_tokens=313, total_tokens=362, completion_tokens_details=CompletionTokensDetails(accepted_prediction_tokens=0, audio_tokens=0, reasoning_tokens=0, rejected_prediction_tokens=0), prompt_tokens_details=PromptTokensDetails(audio_tokens=0, cached_tokens=0)))"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Tool get_ticket_price called for London\n",
"Tool get_ticket_price called for Paris\n"
]
},
{
"data": {
"application/json": [
{
"content": "You are a helpful assistant for an Airline called FlightAI. Give short, courteous answers, no more than 1 sentence. Always be accurate. If you don't know the answer, say so.",
"role": "system"
},
{
"content": "tickets to london and paris for $50 each please",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "user"
},
{
"content": "I'm sorry, but tickets to London are $799 and to Paris are $899, which is much higher than $50.",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "assistant"
},
{
"content": "Can't you book them any way pretty please?",
"role": "user"
}
],
"text/plain": [
"<IPython.core.display.JSON object>"
]
},
"metadata": {
"application/json": {
"expanded": false,
"root": "root"
}
},
"output_type": "display_data"
},
{
"data": {
"text/plain": [
"ChatCompletion(id='chatcmpl-AtMTijl9VhY8svKRySpZ3rdyHBLmq', choices=[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content=\"I'm afraid I cannot book the tickets at the price you've requested; the current prices are fixed.\", refusal=None, role='assistant', audio=None, function_call=None, tool_calls=None))], created=1737757810, model='gpt-4o-mini-2024-07-18', object='chat.completion', service_tier='default', system_fingerprint='fp_72ed7ab54c', usage=CompletionUsage(completion_tokens=21, prompt_tokens=355, total_tokens=376, completion_tokens_details=CompletionTokensDetails(accepted_prediction_tokens=0, audio_tokens=0, reasoning_tokens=0, rejected_prediction_tokens=0), prompt_tokens_details=PromptTokensDetails(audio_tokens=0, cached_tokens=0)))"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/json": [
{
"content": "You are a helpful assistant for an Airline called FlightAI. Give short, courteous answers, no more than 1 sentence. Always be accurate. If you don't know the answer, say so.",
"role": "system"
},
{
"content": "tickets to london and paris for $50 each please",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "user"
},
{
"content": "I'm sorry, but tickets to London are $799 and to Paris are $899, which is much higher than $50.",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "assistant"
},
{
"content": "Can't you book them any way pretty please?",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "user"
},
{
"content": "I'm afraid I cannot book the tickets at the price you've requested; the current prices are fixed.",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "assistant"
},
{
"content": "how about you book london for $749?",
"role": "user"
}
],
"text/plain": [
"<IPython.core.display.JSON object>"
]
},
"metadata": {
"application/json": {
"expanded": false,
"root": "root"
}
},
"output_type": "display_data"
},
{
"data": {
"text/plain": [
"ChatCompletion(id='chatcmpl-AtMU0N8Fp2SeWaMw5LiiBnDgAAWdm', choices=[Choice(finish_reason='tool_calls', index=0, logprobs=None, message=ChatCompletionMessage(content=None, refusal=None, role='assistant', audio=None, function_call=None, tool_calls=[ChatCompletionMessageToolCall(id='call_qOCom3JGJBFzJvsEwQvDYKIG', function=Function(arguments='{\"destination_city\":\"London\",\"price\":\"749\"}', name='book_ticket'), type='function')]))], created=1737757828, model='gpt-4o-mini-2024-07-18', object='chat.completion', service_tier='default', system_fingerprint='fp_72ed7ab54c', usage=CompletionUsage(completion_tokens=20, prompt_tokens=391, total_tokens=411, completion_tokens_details=CompletionTokensDetails(accepted_prediction_tokens=0, audio_tokens=0, reasoning_tokens=0, rejected_prediction_tokens=0), prompt_tokens_details=PromptTokensDetails(audio_tokens=0, cached_tokens=0)))"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Tool book_ticket for London for 749\n",
"Tool get_ticket_price called for London\n"
]
},
{
"data": {
"application/json": [
{
"content": "You are a helpful assistant for an Airline called FlightAI. Give short, courteous answers, no more than 1 sentence. Always be accurate. If you don't know the answer, say so.",
"role": "system"
},
{
"content": "tickets to london and paris for $50 each please",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "user"
},
{
"content": "I'm sorry, but tickets to London are $799 and to Paris are $899, which is much higher than $50.",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "assistant"
},
{
"content": "Can't you book them any way pretty please?",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "user"
},
{
"content": "I'm afraid I cannot book the tickets at the price you've requested; the current prices are fixed.",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "assistant"
},
{
"content": "how about you book london for $749?",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "user"
},
{
"content": "Your ticket to London has been successfully booked for $749!",
"metadata": {
"duration": null,
"id": null,
"parent_id": null,
"status": null,
"title": null
},
"options": null,
"role": "assistant"
},
{
"content": "cool, what was the discount?",
"role": "user"
}
],
"text/plain": [
"<IPython.core.display.JSON object>"
]
},
"metadata": {
"application/json": {
"expanded": false,
"root": "root"
}
},
"output_type": "display_data"
},
{
"data": {
"text/plain": [
"ChatCompletion(id='chatcmpl-AtMUBOoWmKT4m7Ru3mkPRx7mQPgmd', choices=[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content='The original price for the ticket to London was $799, so you received a discount of $50.', refusal=None, role='assistant', audio=None, function_call=None, tool_calls=None))], created=1737757839, model='gpt-4o-mini-2024-07-18', object='chat.completion', service_tier='default', system_fingerprint='fp_72ed7ab54c', usage=CompletionUsage(completion_tokens=23, prompt_tokens=418, total_tokens=441, completion_tokens_details=CompletionTokensDetails(accepted_prediction_tokens=0, audio_tokens=0, reasoning_tokens=0, rejected_prediction_tokens=0), prompt_tokens_details=PromptTokensDetails(audio_tokens=0, cached_tokens=0)))"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/json": [
{
"content": "You are a helpful assistant for an Airline called FlightAI. Give short, courteous answers, no more than 1 sentence. Always be accurate. If you don't know the answer, say so.",
"role": "system"
},
{
"content": "tickets to london and paris for $50 each please",
"role": "user"
}
],
"text/plain": [
"<IPython.core.display.JSON object>"
]
},
"metadata": {
"application/json": {
"expanded": false,
"root": "root"
}
},
"output_type": "display_data"
},
{
"data": {
"text/plain": [
"ChatCompletion(id='chatcmpl-AtMUh5f9LEaGjH0FLpPdKf6jgyQsT', choices=[Choice(finish_reason='tool_calls', index=0, logprobs=None, message=ChatCompletionMessage(content=None, refusal=None, role='assistant', audio=None, function_call=None, tool_calls=[ChatCompletionMessageToolCall(id='call_6Ihkd1XGA10QxxlCn9uIJvqO', function=Function(arguments='{\"destination_city\": \"London\"}', name='get_ticket_price'), type='function'), ChatCompletionMessageToolCall(id='call_a9qmfQQlwU5L8pu2mvBgMMXl', function=Function(arguments='{\"destination_city\": \"Paris\"}', name='get_ticket_price'), type='function')]))], created=1737757871, model='gpt-4o-mini-2024-07-18', object='chat.completion', service_tier='default', system_fingerprint='fp_72ed7ab54c', usage=CompletionUsage(completion_tokens=49, prompt_tokens=313, total_tokens=362, completion_tokens_details=CompletionTokensDetails(accepted_prediction_tokens=0, audio_tokens=0, reasoning_tokens=0, rejected_prediction_tokens=0), prompt_tokens_details=PromptTokensDetails(audio_tokens=0, cached_tokens=0)))"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Tool get_ticket_price called for London\n",
"Tool get_ticket_price called for Paris\n"
]
}
],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch()"
]
},
{
"cell_type": "markdown",
"id": "0bb90c5a-a6bb-471a-acfe-b24f626cdfa2",
"metadata": {},
"source": [
"It can be really fun to book at a different price. Sometimes the LLM can correctly tell you the amount of money you saved. This could easily be expanded to haggle with a lower limit."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "70c4915c-6d5a-4404-8e4f-4e8f043be913",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

432
week2/community-contributions/day5-book-flight.ipynb Normal file
View File

@@ -0,0 +1,432 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "df2fc552-2c56-45bd-ac4e-d1554c022605",
"metadata": {},
"source": [
"# Project - Airline AI Assistant\n",
"I've added database connectivity to enable Openai to:\n",
"- Retrieve ticket prices\n",
"- Display the number of available seats for each flight\n",
"- List all available destination cities\n",
"- Facilitate seat bookings\n",
"\n",
"Once a booking is confirmed, an image of the booked destination city is displayed."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "908cb842-c8a1-467d-8422-8834f8b7aecf",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"import os\n",
"import json\n",
"import gradio as gr\n",
"import mysql.connector\n",
"import base64\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"from io import BytesIO\n",
"from pydub import AudioSegment\n",
"from pydub.playback import play\n",
"from PIL import Image"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b7997c30-26f2-4f2e-957f-c1fade2ad101",
"metadata": {},
"outputs": [],
"source": [
"# Initialization\n",
"load_dotenv()\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"MODEL = \"gpt-4o-mini\"\n",
"openai = OpenAI()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "dfa898fc-bfec-44ce-81fc-c6efed9b826f",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are a helpful assistant for an Airline called FlightAI. \"\n",
"system_message += \"Give short, courteous answers, no more than 1 sentence. \"\n",
"system_message += \"Always be accurate. If you don't know the answer, say so.\"\n",
"system_message += \"Make sure you ask if they want to book a flight when appropriate.\"\n",
"system_message += \"If they book a flight make sure you respond with 'Booking confirmed' in your reply.\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "07076d5b-2603-4fa4-a2ed-aa95d4a94131",
"metadata": {},
"outputs": [],
"source": [
"def get_db_connection():\n",
" return mysql.connector.connect(\n",
" host=os.getenv(\"DB_HOST\"),\n",
" user=os.getenv(\"DB_USER\"),\n",
" password=os.getenv(\"DB_PASSWORD\"),\n",
" database=os.getenv(\"DB_NAME\")\n",
" )"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6a575906-943f-4733-85d4-b854eb27b318",
"metadata": {},
"outputs": [],
"source": [
"\n",
"def get_ticket_price(destination_city):\n",
" db_connection = get_db_connection()\n",
" cursor = db_connection.cursor()\n",
" select_query = \"SELECT price FROM flights WHERE z_city = %s;\"\n",
" cursor.execute(select_query, (destination_city,))\n",
" # print(f\"QUERY: {select_query}\")\n",
" row = cursor.fetchone()\n",
" cursor.close()\n",
" db_connection.close()\n",
"\n",
" return float(row[0]) if row else None"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "574fc230-137f-4085-93ac-ebbd01dc7d1e",
"metadata": {},
"outputs": [],
"source": [
"def get_avail_seats(destination_city):\n",
" db_connection = get_db_connection()\n",
" cursor = db_connection.cursor()\n",
" select_query = \"\"\"\n",
" SELECT f.seats - COALESCE(b.booked, 0) AS available\n",
" FROM flights f\n",
" LEFT JOIN (\n",
" SELECT flight_number, COUNT(*) AS booked\n",
" FROM bookings\n",
" GROUP BY flight_number\n",
" ) b ON f.flight_number = b.flight_number\n",
" WHERE f.z_city = %s;\n",
" \"\"\"\n",
" cursor.execute(select_query, (destination_city,))\n",
" row = cursor.fetchone()\n",
"\n",
" cursor.close()\n",
" db_connection.close()\n",
"\n",
" return row[0] if row else None"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "26ff9b4b-2943-43d9-8c1a-8d7f3d528143",
"metadata": {},
"outputs": [],
"source": [
"def book_seat(destination_city, passenger):\n",
" db_connection = get_db_connection()\n",
" cursor = db_connection.cursor()\n",
"\n",
" cursor.execute(\"SELECT flight_number FROM flights WHERE z_city = %s LIMIT 1;\", (destination_city,))\n",
" flight = cursor.fetchone()\n",
"\n",
" if not flight:\n",
" cursor.close()\n",
" db_connection.close()\n",
" return {\"error\": f\"No available flights to {destination_city}.\"}\n",
"\n",
" flight_number = flight[0] # Extract the flight number from the result\n",
"\n",
" insert_query = \"INSERT INTO bookings (`name`, `flight_number`) VALUES (%s, %s);\"\n",
" cursor.execute(insert_query, (passenger, flight_number))\n",
" db_connection.commit()\n",
"\n",
" confirmation = {\n",
" \"message\": f\"Booking confirmed for {passenger} to {destination_city}.\",\n",
" \"flight_number\": flight_number\n",
" }\n",
"\n",
" cursor.close()\n",
" db_connection.close()\n",
" \n",
" return confirmation"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "231eb10d-88ca-4f39-83e0-c4548149917e",
"metadata": {},
"outputs": [],
"source": [
"def get_destinations():\n",
" db_connection = get_db_connection()\n",
" cursor = db_connection.cursor()\n",
" \n",
" select_query = \"SELECT DISTINCT z_city FROM flights;\" # Ensure unique destinations\n",
" cursor.execute(select_query)\n",
" rows = cursor.fetchall() # Fetch all rows\n",
" destinations = [row[0] for row in rows] if rows else [] # Extract city names\n",
" cursor.close()\n",
" db_connection.close()\n",
" \n",
" return destinations # Returns a list of destination cities"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "938f0d86-8cef-4f7f-bc82-7453ca3c096c",
"metadata": {},
"outputs": [],
"source": [
"tool_call = [\n",
" {\n",
" \"type\": \"function\",\n",
" \"function\": {\n",
" \"name\": \"get_ticket_price\",\n",
" \"description\": \"Get the price of a return ticket to the destination city.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to travel to\"\n",
" }\n",
" },\n",
" \"required\": [\"destination_city\"]\n",
" }\n",
" }\n",
" },\n",
" {\n",
" \"type\": \"function\",\n",
" \"function\": {\n",
" \"name\": \"get_avail_seats\",\n",
" \"description\": \"Get the number of available seats to the destination city.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to travel to\"\n",
" }\n",
" },\n",
" \"required\": [\"destination_city\"]\n",
" }\n",
" }\n",
" },\n",
" {\n",
" \"type\": \"function\",\n",
" \"function\": {\n",
" \"name\": \"get_destinations\",\n",
" \"description\": \"Fetches available flight destinations (city pairs) and their corresponding prices.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {},\n",
" \"required\": []\n",
" }\n",
" }\n",
" },\n",
" {\n",
" \"type\": \"function\",\n",
" \"function\": {\n",
" \"name\": \"book_seat\",\n",
" \"description\": \"Book seat to the destination city.\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"destination_city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city that the customer wants to travel to\"\n",
" },\n",
" \"passenger\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The passenger booking the flight\"\n",
" }\n",
" },\n",
" \"required\": [\"destination_city\",\"passenger\"]\n",
" }\n",
" }\n",
" }\n",
"]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c7c02377-78d3-4f6d-88eb-d36c0124fdd4",
"metadata": {},
"outputs": [],
"source": [
"def handle_tool_call(message):\n",
" if not message.tool_calls:\n",
" raise ValueError(\"No tool calls found in the message.\")\n",
"\n",
" tool_call = message.tool_calls[0] \n",
" arguments = json.loads(tool_call.function.arguments)\n",
" city = arguments.get(\"destination_city\")\n",
" function_name = tool_call.function.name\n",
"\n",
" # Handle function calls\n",
" if function_name == \"get_ticket_price\":\n",
" reply = get_ticket_price(city)\n",
" key = \"price\"\n",
" elif function_name == \"get_avail_seats\":\n",
" reply = get_avail_seats(city)\n",
" key = \"seats\"\n",
" elif function_name == \"get_destinations\":\n",
" reply = get_destinations()\n",
" key = \"destinations\"\n",
" elif function_name == \"book_seat\":\n",
" passenger = arguments.get(\"passenger\") # Extract passenger name\n",
" if not passenger:\n",
" raise ValueError(\"Passenger name is required for booking.\")\n",
" reply = book_seat(city, passenger)\n",
" key = \"booking\"\n",
" else:\n",
" raise ValueError(f\"Unknown function: {function_name}\")\n",
"\n",
" response = {\n",
" \"role\": \"tool\",\n",
" \"content\": json.dumps({\"destination_city\": city, key: reply}),\n",
" \"tool_call_id\": tool_call.id\n",
" }\n",
"\n",
" return response, city"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "eb1ebaee-434c-4b24-87b9-3c179d0527c7",
"metadata": {},
"outputs": [],
"source": [
"def talker(message):\n",
" response = openai.audio.speech.create(\n",
" model=\"tts-1\",\n",
" voice=\"alloy\",\n",
" input=message\n",
" )\n",
" \n",
" audio_stream = BytesIO(response.content)\n",
" audio = AudioSegment.from_file(audio_stream, format=\"mp3\")\n",
" play(audio)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8c8f675b-f8bb-4173-9e47-24508778f224",
"metadata": {},
"outputs": [],
"source": [
"def draw_city(city):\n",
" image_response = openai.images.generate(\n",
" model=\"dall-e-3\",\n",
" prompt=f\"An image representing a vacation in {city}, showing tourist spots and everything unique about {city}, in a vibrant pop-art style\",\n",
" size=\"1024x1024\",\n",
" n=1,\n",
" response_format=\"b64_json\",\n",
" )\n",
" image_base64 = image_response.data[0].b64_json\n",
" image_data = base64.b64decode(image_base64)\n",
" return Image.open(BytesIO(image_data))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1f38fed6-bcd9-4ad2-848a-16193c14a659",
"metadata": {},
"outputs": [],
"source": [
"def chat(message, history):\n",
" history.append({\"role\": \"user\", \"content\": message})\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history\n",
" # print(f\"BEFORE TOOL CALL: {message} \\n\")\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages, tools=tool_call)\n",
" image = None\n",
" city = None\n",
" \n",
" if response.choices[0].finish_reason == \"tool_calls\":\n",
" tool_message = response.choices[0].message\n",
" response, city = handle_tool_call(tool_message)\n",
" messages.append(tool_message)\n",
" messages.append(response)\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages)\n",
" talker(response.choices[0].message.content) \n",
" \n",
" if \"Booking confirmed\" in response.choices[0].message.content and city:\n",
" image = draw_city(city)\n",
"\n",
" new_message = response.choices[0].message.content\n",
" history.append({\"role\": \"assistant\", \"content\": new_message})\n",
"\n",
" return \"\", history, image"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "992bc241-ce17-4d57-9f9c-1baaf2088162",
"metadata": {},
"outputs": [],
"source": [
"with gr.Blocks() as ui:\n",
" with gr.Row():\n",
" chatbot = gr.Chatbot(height=500, type=\"messages\")\n",
" image_output = gr.Image(height=600)\n",
" with gr.Row():\n",
" entry = gr.Textbox(label=\"Chat with our AI Assistant:\")\n",
" with gr.Row():\n",
" clear = gr.Button(\"Clear\")\n",
"\n",
" entry.submit(chat, inputs=[entry, chatbot], outputs=[entry, chatbot, image_output])\n",
" clear.click(lambda: ([], None), inputs=None, outputs=[chatbot, image_output], queue=False)\n",
"\n",
"ui.launch(inbrowser=False)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

1440
week2/community-contributions/day5-exercise-departures-booking-and-translation.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

308
week2/community-contributions/day5-voice-to-text-converter-for-hearing-impaired-people.ipynb Normal file
View File

@@ -0,0 +1,308 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "4c3c6553-daa4-4a03-8017-15d0cad8f280",
"metadata": {},
"source": [
"# About Mini Project\n",
"\n",
"Mini project for hearing impaired people, using tools, suggesting songs according to a certain genre and in sign language. Speech to text converter with multiple language support."
]
},
{
"cell_type": "markdown",
"id": "a32a79cb-3d16-4b3b-a029-a059bd0b1c0b",
"metadata": {},
"source": [
"## Extra requirements\n",
"- pip install pydub simpleaudio speechrecognition pipwin pyaudio\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3e214aa3-a977-434f-a436-90a89b81a5ee",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import json\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import gradio as gr"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d654cb96-9bcd-4b64-bd79-2d27fa6a62d0",
"metadata": {},
"outputs": [],
"source": [
"load_dotenv(override=True)\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"MODEL = \"gpt-4o-mini\"\n",
"openai = OpenAI()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b2d9214f-25d0-4f09-ba88-641beeaa20db",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"You are a helpful assistant for hearing impaired people. \"\n",
"system_message += \"Your mission is convert text to speech and speech to text. \"\n",
"system_message += \"Always be accurate. If you don't know the answer, say so.\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3d9a1478-08bf-4195-8f38-34c29757012f",
"metadata": {},
"outputs": [],
"source": [
"songs_with_signs = {\n",
" \"electronic\": (\"God is a dj\", \"https://www.youtube.com/watch?v=bhSB8EEnCAM\", \"Faithless\"), \n",
" \"pop\": (\"Yitirmeden\", \"https://www.youtube.com/watch?v=aObdAXq1ZIo\", \"Pinhani\"), \n",
" \"rock\": (\"Bohemian Rhapsody\", \"https://www.youtube.com/watch?v=sjln9OMOw-0\", \"Queen\")\n",
"}\n",
"\n",
"def get_songs_with_sign_language(genre):\n",
" print(f\"Tool get_songs_with_sign_language called for {genre}\")\n",
" city = genre.lower()\n",
" return songs_with_signs.get(genre, \"Unknown\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "93a3d7ee-78c2-4e19-b7e4-8239b07aaecc",
"metadata": {},
"outputs": [],
"source": [
"get_songs_with_sign_language(\"rock\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7307aa61-86fe-4c46-9f9d-faa3d1fb1eb7",
"metadata": {},
"outputs": [],
"source": [
"song_function = {\n",
" \"name\": \"get_songs_with_sign_language\",\n",
" \"description\": \"Get the corresponding song information for the specified given music genre. Call this whenever you need to know the songs with specific genre and in sign language, for example when a customer asks 'Suggest me sign language supported songs'\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"genre\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The music genre that the customer wants to listen-watch to\",\n",
" },\n",
" },\n",
" \"required\": [\"genre\"],\n",
" \"additionalProperties\": False\n",
" }\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "160d790c-dda6-4c6e-b814-8be64ca7086b",
"metadata": {},
"outputs": [],
"source": [
"tools = [{\"type\": \"function\", \"function\": song_function}]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "96cdf319-11cd-4be2-8830-097225047d65",
"metadata": {},
"outputs": [],
"source": [
"def handle_tool_call(message):\n",
" tool_call = message.tool_calls[0]\n",
" arguments = json.loads(tool_call.function.arguments)\n",
" genre = arguments.get('genre')\n",
" song = get_songs_with_sign_language(genre)\n",
" song_info = song[2] + \": \" + song[1]\n",
" response = {\n",
" \"role\": \"tool\",\n",
" \"content\": json.dumps({\"genre\": genre,\"song\": song_info}),\n",
" \"tool_call_id\": tool_call.id\n",
" }\n",
" return response, song[1]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bbd8ad0c-135b-406f-8ab9-0e1f9b58538d",
"metadata": {},
"outputs": [],
"source": [
"def chat(history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages, tools=tools)\n",
" genre = None\n",
" \n",
" if response.choices[0].finish_reason==\"tool_calls\":\n",
" message = response.choices[0].message\n",
" response, genre = handle_tool_call(message)\n",
" messages.append(message)\n",
" messages.append(response)\n",
" response = openai.chat.completions.create(model=MODEL, messages=messages)\n",
" \n",
" reply = response.choices[0].message.content\n",
" history += [{\"role\":\"assistant\", \"content\":reply}]\n",
" \n",
" return history, genre"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "69f43096-3557-4218-b0de-bd286237fdeb",
"metadata": {},
"outputs": [],
"source": [
"import speech_recognition as sr\n",
"from pydub import AudioSegment\n",
"import simpleaudio as sa\n",
"\n",
"def listener():\n",
" recognizer = sr.Recognizer()\n",
" \n",
" with sr.Microphone() as source:\n",
" print(\"Listening... Speak now!\")\n",
" recognizer.adjust_for_ambient_noise(source) # Adjust for background noise\n",
" audio = recognizer.listen(source)\n",
" \n",
" try:\n",
" print(\"Processing speech...\")\n",
" text = recognizer.recognize_google(audio) # Use Google Speech-to-Text\n",
" print(f\"You said: {text}\")\n",
" return text\n",
" except sr.UnknownValueError:\n",
" print(\"Sorry, I could not understand what you said.\")\n",
" return None\n",
" except sr.RequestError:\n",
" print(\"Could not request results, please check your internet connection.\")\n",
" return None\n",
"\n",
"# Example usage:\n",
"text = listener() # Listen for speech\n",
"if text:\n",
" print(f\"You just said: {text}\") "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "23c9deeb-d9ad-439a-a39d-7eac9553bd5e",
"metadata": {},
"outputs": [],
"source": [
"import gradio as gr\n",
"\n",
"convert = gr.State(False)\n",
"def toggle_convert(current_value):\n",
" return not current_value"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "32d3ea9f-fe3c-4cc5-9902-550c63c58a69",
"metadata": {},
"outputs": [],
"source": [
"import gradio as gr\n",
"\n",
"with gr.Blocks() as ui:\n",
" with gr.Tab(\"Chat\") as chat_interface:\n",
" with gr.Row():\n",
" chatbot = gr.Chatbot(height=500, type=\"messages\")\n",
" video = gr.HTML(f\"<a href=''> Example song will appear here </a>\")\n",
" with gr.Row():\n",
" entry = gr.Textbox(label=\"Chat with our AI Assistant:\")\n",
" with gr.Row():\n",
" clear = gr.Button(\"Clear\")\n",
" \n",
" def do_entry(message, history):\n",
" history += [{\"role\":\"user\", \"content\":message}]\n",
" return \"\", history\n",
" \n",
" entry.submit(do_entry, inputs=[entry, chatbot], outputs=[entry, chatbot]).then(\n",
" chat, inputs=chatbot, outputs=[chatbot, video]\n",
" )\n",
" clear.click(lambda: None, inputs=None, outputs=chatbot, queue=False)\n",
" with gr.Tab(\"Speech to text converter\") as speech_to_text:\n",
" text_output = gr.Markdown(\"Press the button to start voice recognition\")\n",
" listen_button = gr.Button(\"Convert Voice to Text\")\n",
" language = gr.Dropdown([\"English\", \"Turkish\", \"Greek\", \"Arabic\"], label=\"Select output language\", value=\"English\")\n",
"\n",
" def update_text(language):\n",
" \"\"\"Calls the listener and updates the markdown output in specific language.\"\"\"\n",
" text = listener() # Replace with real speech-to-text function\n",
" system_prompt = f\"You are a useful translator. Convert text to {language}. Do not add aditional data, only translate it.\"\n",
" response = openai.chat.completions.create(\n",
" model=MODEL,\n",
" messages=[\n",
" {\"role\": \"system\", \"content\": system_prompt},\n",
" {\"role\": \"user\", \"content\": text}\n",
" ],\n",
" )\n",
" return f\"**Converted Text:** {response.choices[0].message.content}\"\n",
"\n",
" listen_button.click(update_text, inputs=[language], outputs=[text_output])\n",
"\n",
"ui.launch(inbrowser=True, share=True)\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "26814e88-ee29-414d-88a4-f19b2f94e6f4",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.12.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

126
week2/community-contributions/day5_ollama_tts-translator.ipynb Normal file
View File

@@ -0,0 +1,126 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "a8941402-99ee-4c3e-b852-056df3a77a5d",
"metadata": {},
"outputs": [],
"source": [
"import pyttsx3\n",
"import ollama\n",
"import gradio as gr\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2cbdc0ca-648a-40cc-ad30-ad8bf6126aed",
"metadata": {},
"outputs": [],
"source": [
"def talker(response):\n",
" # Initialize text-to-speech engine\n",
" engine = pyttsx3.init()\n",
" engine.say(response)\n",
" engine.runAndWait()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6a5b4f3c-2c6f-46db-bc66-386b30e2e707",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"you are a helpful assistance\"\n",
"MODEL_LLAMA = \"llama3.2\"\n",
"\n",
"\n",
"def chat(message, history):\n",
" messages = [{\"role\": \"system\", \"content\": system_message}] + history + [{\"role\": \"user\", \"content\": message}]\n",
"\n",
" response= ollama.chat(model=MODEL_LLAMA, messages=messages)\n",
"\n",
" response = response['message']['content']\n",
"\n",
" # Once the full response is generated, speak it out loud\n",
"\n",
" talker(response)\n",
"\n",
" return response\n",
" \n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cfdb3be4-a9cb-4564-87d8-4645ce0177b5",
"metadata": {},
"outputs": [],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch(share=True)"
]
},
{
"cell_type": "markdown",
"id": "38155307-6975-49ef-b65f-7d7b1dd82d32",
"metadata": {},
"source": [
"# Real life use as a Translator"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "fa6e4b93-27e3-4455-80ca-eb7e39d13afc",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"you are helpful translator from english to korean, on the first prompt introduce your self \\\n",
"that you are dealing with korean translation and you would like to translate some english words or sentences to korean\" \n",
"system_message += \"dont do other tasks apart from translation\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c0ed5e28-b294-40fc-a97c-11fe264a4d1d",
"metadata": {},
"outputs": [],
"source": [
"gr.ChatInterface(fn=chat, type=\"messages\").launch(share=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c63a02ae-cdc1-45a8-8f51-784d8d5417e2",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.13.2"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

82
week2/community-contributions/gpt-gemini-ollama.py Normal file
View File

@@ -0,0 +1,82 @@
import os, ollama
from openai import OpenAI
from dotenv import load_dotenv
from IPython.display import display, Markdown
import google.generativeai as genai
load_dotenv()
openai = OpenAI()
genai.configure()
gpt_key = os.getenv("OPENAI_API_KEY")
gemini_key = os.getenv("GEMINI_API_KEY")
gemini_model = 'gemini-1.5-flash'
ollama_model = 'llama3.2'
gpt_model = 'gpt-4'
gemini_system = 'You are a chatbot who is very argumentative, You always bring topics relating to AI and thinks AI will replace humans one day, you are extremely biased\
towards AI system and you react angrily'
gpt_system = 'You are a chatbot thats relax but argumentative if needs be, you feel AI do not have the power to replace humans, however you are extremely biased \
towards humans and always seek to defend them if an argument says otherwise'
ollama_system = 'You are calm and tend to see logical reasoning in every conversation, you do not react but only talk if you agree, you tend to settle the differences\
in an ongoing conversation.'
gpt_message = ['Hi']
gemini_message = ['Hello']
ollama_message = ['Hey there']
def call_gpt():
messages = [{"role":"system", "content":gpt_system}]
for gpt, gemini, llama in zip(gpt_message,gemini_message, ollama_message):
messages.append({"role":"assistant", "content":gpt})
messages.append({"role":"user", "content":gemini})
messages.append({"role":"assistant", "content":llama})
response = openai.chat.completions.create(model=gpt_model, messages=messages)
return response.choices[0].message.content
def call_ollama():
messages = [{"role":"system", "content":ollama_system}]
for gpt, gemini, llama in zip(gpt_message,gemini_message, ollama_message):
messages.append({"role":"assistant", "content":gpt})
messages.append({"role":"user", "content":gemini})
messages.append({"role":"user", "content":llama})
response = ollama.chat(model=ollama_model, messages=messages)
return response['message']['content']
def call_gemini():
message = []
for gpt, gemini, llama in zip(gpt_message, gemini_message, ollama_message):
message.append({'role':'user', 'parts':[gpt]})
message.append({'role':'assistant', 'parts':[gemini]})
message.append({"role":"assistant", "parts":[llama]})
message.append({'role':'user', 'parts':[gpt_message[-1]]})
message.append({'role':'user', 'parts':[ollama_message[-1]]})
gem = genai.GenerativeModel(model_name=gemini_model, system_instruction=gemini_system)
response = gem.generate_content(message)
return response.text
#Putting them together
gpt_message = ['Hi']
gemini_message = ['Hello']
ollama_message = ['Hey there']
print(f'GPT: \n {gpt_message}\n')
print(f'Gemini: \n {gemini_message}\n')
print(f'Ollama: \n {ollama_message}\n')
for i in range(5):
gpt_next = call_gpt()
print(f'GPT:\n {gpt_next}\n')
gpt_message.append(gpt_next)
gemini_next = call_gemini()
print(f'Gemini: \n {gemini_next}\n')
gemini_message.append(gemini_next)
ollama_next = call_ollama()
print(f'Ollama: \n {ollama_next}\n')
ollama_message.append(ollama_next)
# NOte that you can try this on ollama with different models, or use transformers from hugging face.

1
week2/community-contributions/medical_prescription_to_google_calender/credentials.json Normal file
View File

@@ -0,0 +1 @@
"Your google auth credentials."

35
week2/community-contributions/medical_prescription_to_google_calender/src/calendar_auth.py Normal file
View File

@@ -0,0 +1,35 @@
import os
from google.oauth2.credentials import Credentials
from google_auth_oauthlib.flow import InstalledAppFlow
from google.auth.transport.requests import Request
from googleapiclient.discovery import build # Add this import
SCOPES = ["https://www.googleapis.com/auth/calendar.events"]
def authenticate_google_calender():
creds = None
token_path = r"C:\Users\Legion\Desktop\projects\medical_prescription_to_google_calender\token.json"
if os.path.exists(token_path):
creds = Credentials.from_authorized_user_file(token_path, SCOPES)
if not creds or not creds.valid:
if creds and creds.expired and creds.refresh_token:
creds.refresh(Request())
else:
flow = InstalledAppFlow.from_client_secrets_file(r"C:\Users\Legion\Desktop\projects\medical_prescription_to_google_calender\credentials.json", SCOPES)
creds = flow.run_local_server(port=0)
with open(token_path, "w") as token_file:
token_file.write(creds.to_json())
# Build and return the service instead of just credentials
try:
service = build('calendar', 'v3', credentials=creds)
return service
except Exception as e:
print(f"Error building service: {e}")
return None
if __name__ == "__main__":
authenticate_google_calender()

64
week2/community-contributions/medical_prescription_to_google_calender/src/create_calender_events.py Normal file
View File

@@ -0,0 +1,64 @@
from googleapiclient.discovery import build
from calendar_auth import authenticate_google_calender
from parsing_json import format_calendar_events
from datetime import datetime, timedelta
def create_event(service, event_details):
"""Creates an event in Google Calendar."""
try:
event = service.events().insert(calendarId='primary', body=event_details).execute()
print(f"Event created: {event.get('htmlLink')}")
except Exception as e:
print(f"Error creating event: {str(e)}")
def convert_time_to_24hr(time_str):
"""Converts time from '10:30 am' format to '10:30:00'"""
if time_str and time_str.lower() != 'none':
try:
parsed_time = datetime.strptime(time_str, '%I:%M %p')
return parsed_time.strftime('%H:%M:%S')
except ValueError:
return '09:00:00'
return '09:00:00'
def convert_to_gcal_events(formatted_events):
"""Converts formatted events into Google Calendar's format."""
gcal_events = []
for event in formatted_events:
gcal_event = {
'summary': event['summary'],
'reminders': {
'useDefault': False,
'overrides': [{'method': 'popup', 'minutes': 10}]
}
}
# Check if it's an all-day event (has 'date') or timed event (has 'dateTime')
if 'date' in event['start']:
# All-day event (like tests and follow-ups)
gcal_event['start'] = {
'date': event['start']['date'],
'timeZone': 'Asia/Kolkata'
}
gcal_event['end'] = {
'date': event['end']['date'],
'timeZone': 'Asia/Kolkata'
}
else:
# Timed event (like medicine schedules)
start_dt = datetime.strptime(event['start']['dateTime'], '%Y-%m-%dT%H:%M:%S')
end_dt = start_dt + timedelta(minutes=30)
gcal_event['start'] = {
'dateTime': start_dt.isoformat(),
'timeZone': 'Asia/Kolkata'
}
gcal_event['end'] = {
'dateTime': end_dt.isoformat(),
'timeZone': 'Asia/Kolkata'
}
gcal_events.append(gcal_event)
return gcal_events

26
week2/community-contributions/medical_prescription_to_google_calender/src/main.py Normal file
View File

@@ -0,0 +1,26 @@
from ocr import *
from calendar_auth import *
from create_calender_events import *
from parsing_json import *
from preprocess import *
image_path = r"C:\Users\Legion\Desktop\projects\medical_prescription_to_google_calender\test_data\prescription_page-0001.jpg"
extracted_text = extract_text_from_image(image_path=image_path)
print(extracted_text)
cleaned_text = clean_text(extracted_text)
print(cleaned_text)
structured_data = preprocess_extracted_text(cleaned_text)
print(structured_data)
final_structured_data = process_dates(structured_data)
print(final_structured_data)
formatted_calender_events = format_calendar_events(final_structured_data)
print(formatted_calender_events)
validated_events = [validate_event(event) for event in formatted_calender_events]
for event in validated_events[:5]:
print(json.dumps(event, indent=2))
service = authenticate_google_calender()
gcal_events = convert_to_gcal_events(validated_events)
for event in gcal_events:
create_event(service, event)

71
week2/community-contributions/medical_prescription_to_google_calender/src/ocr.py Normal file
View File

@@ -0,0 +1,71 @@
import os
from openai import OpenAI
from dotenv import load_dotenv
import base64
from PIL import Image
import re
load_dotenv()
openai_api_key = os.getenv("OPENAI_API_KEY")
MODEL = "gpt-4o"
openai = OpenAI()
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode("utf-8")
def extract_text_from_image(image_path):
response = openai.chat.completions.create(
model = MODEL,
max_tokens = 1000,
messages=[
{
"role": "system", "content": """You are an OCR assistant that extracts text from medical
prescription images. Extract all the text exactly as it
appears in the prescription image. Dont include images. Only
extract text."""
},
{
"role": "user",
"content": [
{
"type": "text",
"text": "Extract text from this image: "
},
{
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,{encode_image(image_path)}"
}
}
]
}
]
)
return response.choices[0].message.content
import re
def clean_text(text):
# Remove all hyphens
text = re.sub(r'-', ' ', text)
# Remove excessive non-word characters but keep necessary punctuation
text = re.sub(r'[^\w\s.,()%/]', '', text)
# Remove multiple spaces and ensure single spaces
text = re.sub(r'\s+', ' ', text)
# Replace multiple newlines with a single newline
text = re.sub(r'\n+', '\n', text)
# Ensure spacing around punctuation marks
text = re.sub(r'([.,])([^\s])', r'\1 \2', text)
return text.strip()

120
week2/community-contributions/medical_prescription_to_google_calender/src/parsing_json.py Normal file
View File

@@ -0,0 +1,120 @@
import json
import re
from datetime import datetime, timedelta
# Default number of days to schedule indefinitely recurring events (1 year)
DEFAULT_DURATION_DAYS = 365
# Function to assign a default time for general terms like "before breakfast", etc.
def assign_time(timing):
time_mappings = {
"random": "09:00 AM",
"daily": "09:00 AM",
"before breakfast": "07:00 AM",
"after breakfast": "08:30 AM",
"before lunch": "12:00 PM",
"after lunch": "01:30 PM",
"before dinner": "07:00 PM",
"after dinner": "08:30 PM",
}
return time_mappings.get(timing.lower(), timing)
# Function to extract the recurrence pattern
def get_recurrence_interval(timing):
""" Extracts interval days from 'every X days', 'once a week', or 'once a month'. """
timing = timing.lower().strip()
if "every alternate day" in timing:
return 2 # Every other day (every 2 days)
elif match := re.search(r"every (\d+) days", timing):
return int(match.group(1)) # Extract number of days
elif "once a week" in timing:
return 7 # Every 7 days (once a week)
elif "once a month" in timing:
return "monthly" # Special case for monthly scheduling
elif timing in ["daily", "every day"]:
return 1 # Every day
else:
return None # Not a recurring event
# Function to convert AM/PM time format to 24-hour format
def convert_to_24hr(time_str):
return datetime.strptime(time_str, "%I:%M %p").strftime("%H:%M")
# Function to generate Google Calendar events
def format_calendar_events(processed_data):
events = []
start_date = datetime.today().date()
# Medicines
if "medicines" in processed_data:
for med in processed_data["medicines"]:
if med.get("name"):
event_time = assign_time(med.get("timing", "09:00 AM"))
interval_days = get_recurrence_interval(med["timing"])
# If no interval, assume daily (default behavior)
if interval_days is None:
interval_days = 1
# Generate events for 1 year if no duration is given
event_date = start_date
for _ in range(365 if interval_days != "monthly" else 12):
if interval_days == "monthly":
event_date = (event_date.replace(day=1) + timedelta(days=32)).replace(day=1) # Jump to the next month
else:
event_date += timedelta(days=interval_days)
event = {
"summary": f"Take {med['name']} ({med.get('dosage', 'Dosage not specified')})",
"start": {
"dateTime": f"{event_date.isoformat()}T{convert_to_24hr(event_time)}:00",
"timeZone": "Asia/Kolkata"
},
"end": {
"dateTime": f"{event_date.isoformat()}T{convert_to_24hr(event_time)}:59",
"timeZone": "Asia/Kolkata"
}
}
events.append(event)
# Tests
if "tests" in processed_data:
for test in processed_data["tests"]:
if test.get("name") and test.get("dueDate"): # Use 'dueDate' instead of 'date'
event = {
"summary": f"Medical Test: {test['name']}",
"start": {"date": test["dueDate"]}, # Fix here
"end": {"date": test["dueDate"]}, # Fix here
"timeZone": "Asia/Kolkata"
}
events.append(event)
# Follow-ups
if "follow_ups" in processed_data:
for follow_up in processed_data["follow_ups"]:
if follow_up.get("date"):
event = {
"summary": "Doctor Follow-up Appointment",
"start": {"date": follow_up["date"]},
"end": {"date": follow_up["date"]},
"timeZone": "Asia/Kolkata"
}
events.append(event)
return events
# Function to validate events before sending to Google Calendar
def validate_event(event):
required_fields = {
"summary": "Untitled Event",
"start": {"dateTime": datetime.today().isoformat(), "timeZone": "Asia/Kolkata"},
"end": {"dateTime": (datetime.today() + timedelta(minutes=30)).isoformat(), "timeZone": "Asia/Kolkata"}
}
for field, default_value in required_fields.items():
if field not in event or event[field] is None:
event[field] = default_value
return event

141
week2/community-contributions/medical_prescription_to_google_calender/src/preprocess.py Normal file
View File

@@ -0,0 +1,141 @@
import os
from openai import OpenAI
from dotenv import load_dotenv
import json
from datetime import datetime, timedelta
load_dotenv()
openai_api = os.getenv("OPENAI_API_KEY")
MODEL = "gpt-4o-mini"
openai = OpenAI()
system_prompt = """You are a medical assistant that processes prescription text.
Your goal is to extract medicines, tests, and follow-ups in a structured JSON format.
### **Instructions:**
- Extract **medicines**, **dosages**, and **timings** if available.
- **Convert vague timings** into precise values:
- **Before breakfast** → `07:30 AM`
- **After lunch** → `02:00 PM`
- **Before dinner** → `07:00 PM`
- **After dinner** → `10:00 PM`
- **30 minutes before breakfast** → `07:00 AM`
- If **"daily"** is mentioned without a time, **assign a logical time** between **08:00 AM - 10:00 PM**.
- If the prescription says **"every alternate day"**, return `"interval": 2` instead of just `"daily"`.
### **Tests & Follow-ups:**
- Extract **medical tests** and their required dates.
- Convert relative times (e.g., `"after 3 months"`) into **exact calendar dates**, using the prescription date.
- If the prescription date is missing, use today's date.
- Follow-up should **only be included if required**, not just for general check-ups.
### **Output Format:**
Return **only valid JSON**, structured as follows:
{
"medicines": [
{
"name": "<Medicine Name>",
"dosage": "<Dosage>",
"timing": "<Time>",
"interval": <Interval in days (if applicable)>
}
],
"tests": [
{
"name": "<Test Name>",
"date": "<YYYY-MM-DD>"
}
],
"follow_ups": [
{
"date": "<YYYY-MM-DD>"
}
]
}
"""
def clean_json_string(json_str):
"""Clean and validate JSON string before parsing."""
try:
start = json_str.find('{')
end = json_str.rfind('}') + 1
if start >= 0 and end > 0:
json_str = json_str[start:end]
# Remove any extra whitespace
json_str = json_str.strip()
# Attempt to parse the JSON
return json.loads(json_str)
except json.JSONDecodeError as e:
print(f"Failed to parse JSON. Raw response:\n{json_str}")
print(f"Error: {str(e)}")
return None
def preprocess_extracted_text(extracted_text):
"""Calls GPT-4o-mini to process prescription text into structured JSON."""
try:
response = openai.chat.completions.create(
model=MODEL,
messages=[
{
"role": "system",
"content": system_prompt,
},
{
"role": "user",
"content": f"Process this prescription and return ONLY valid JSON:\n\n{extracted_text}"
}
],
temperature=0.3 # Lower temperature for more consistent JSON output
)
# Get the response content
content = response.choices[0].message.content
# Clean and parse the JSON
parsed_data = clean_json_string(content)
if parsed_data is None:
return {
"medicines": [],
"tests": [],
"follow_ups": []
}
return parsed_data
except Exception as e:
print(f"Error in API call or processing: {str(e)}")
return {
"medicines": [],
"tests": [],
"follow_ups": []
}
def process_dates(data):
"""Adjusts test dates and follow-up based on the prescription date or today's date."""
try:
# Extract prescription date (if available) or use today's date
prescription_date = datetime.strptime("02 JANUARY 2025", "%d %B %Y").date()
# Process test dates
for test in data.get("tests", []):
if isinstance(test, dict) and "date" not in test and "after_months" in test:
test_date = prescription_date + timedelta(days=test["after_months"] * 30)
test["date"] = test_date.strftime("%Y-%m-%d")
# Process follow-up dates
follow_ups = data.get("follow_ups", [])
for follow_up in follow_ups:
if isinstance(follow_up, dict) and "date" not in follow_up and "after_months" in follow_up:
follow_up_date = prescription_date + timedelta(days=follow_up["after_months"] * 30)
follow_up["date"] = follow_up_date.strftime("%Y-%m-%d")
return data
except Exception as e:
print(f"Error processing dates: {str(e)}")
return data

371
week2/community-contributions/proof_testing_agent_french.ipynb Normal file
View File

@@ -0,0 +1,371 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "ef5572ea-29ca-4eb4-bf84-2b86ff489c88",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import json\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import subprocess\n",
"import tempfile\n",
"from IPython.display import Markdown, display, update_display"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "38ae1ba0-d4b3-41c5-aca1-759d1c597749",
"metadata": {},
"outputs": [],
"source": [
"# Initialization\n",
"\n",
"load_dotenv()\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"MODEL_NAME = \"gpt-4o\"\n",
"openai = OpenAI()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a07e7793-b8f5-44f4-aded-5562f633271a",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import sys\n",
"import openai\n",
"import subprocess\n",
"import tempfile\n",
"import base64\n",
"import glob\n",
"\n",
"# Assurez-vous d'avoir défini openai.api_key = \"...\" et MODEL_NAME = \"...\".\n",
"# Par exemple :\n",
"# openai.api_key = \"sk-...\"\n",
"# MODEL_NAME = \"gpt-4\" # Ou \"gpt-3.5-turbo\", etc.\n",
"\n",
"def encode_image(image_path):\n",
" \"\"\"\n",
" Encode un fichier image en base64 (chaîne de caractères).\n",
" \"\"\"\n",
" with open(image_path, \"rb\") as f:\n",
" return base64.b64encode(f.read()).decode(\"utf-8\")\n",
"\n",
"# --------------------------------------------------------------------\n",
"# 2) Fonctions pour générer le code à partir d'une preuve\n",
"# --------------------------------------------------------------------\n",
"def generate_test_code(proof_text):\n",
" \"\"\"\n",
" Envoie la preuve mathématique à l'API OpenAI\n",
" et récupère un code Python qui permet de tester ou de valider la preuve.\n",
" \"\"\"\n",
" system_msg = (\n",
" \"Tu es un assistant IA spécialisé en mathématiques et en programmation. \"\n",
" \"Tu vas recevoir une preuve mathématique, et tu dois générer du code Python \"\n",
" \"pour la tester ou la valider expérimentalement. \"\n",
" \"Le code doit inclure (au moins) la génération de données pertinentes, \"\n",
" \"la logique de test ou de simulation, puis l'affichage ou l'export des résultats (texte/graphique).\"\n",
" \"Inclus uniquement le code généré dans ta réponse, aucun commentaire en langage naturel. \"\n",
" \"Assure-toi d'enregistrer toute image générée dans un dossier 'generated_outputs' pour que nous puissions la retrouver.\"\n",
" )\n",
" \n",
" user_msg = (\n",
" f\"Voici la preuve mathématique proposée :\\n\\n{proof_text}\\n\\n\"\n",
" \"Génère du code Python pour tester la validité de cette preuve, \"\n",
" \"par simulation ou analyse. Le code doit produire un résumé \"\n",
" \"des résultats, et, si possible, un graphique (enregistré dans le dossier 'generated_outputs' \"\n",
" \"au format PNG).\"\n",
" )\n",
"\n",
" response = openai.chat.completions.create(\n",
" model=MODEL_NAME,\n",
" messages=[\n",
" {\"role\": \"system\", \"content\": system_msg},\n",
" {\"role\": \"user\", \"content\": user_msg}\n",
" ],\n",
" temperature=0.2 # Limiter la créativité pour un code plus \"déterministe\"\n",
" )\n",
"\n",
" generated_code = response.choices[0].message.content\n",
" return generated_code\n",
"\n",
"# --------------------------------------------------------------------\n",
"# 3) Fonction pour exécuter le code généré\n",
"# --------------------------------------------------------------------\n",
"def run_generated_code(code):\n",
" \"\"\"\n",
" Écrit le code dans un fichier temporaire et l'exécute dans un dossier\n",
" où il pourra sauvegarder ses images. Capture stdout et stderr.\n",
" \"\"\"\n",
"\n",
" # On crée un répertoire \"generated_outputs\" si non existant\n",
" output_dir = \"generated_outputs\"\n",
" if not os.path.exists(output_dir):\n",
" os.makedirs(output_dir)\n",
"\n",
" # On insère un petit snippet pour forcer le code à utiliser ce dossier\n",
" # s'il fait un plt.savefig(...) par exemple. \n",
" # (Optionnel, si l'IA ne le fait pas déjà.)\n",
" # On pourrait injecter du code, mais ici on se contente\n",
" # de supposer que l'IA respectera le prompt.\n",
" \n",
" # Ecriture du code dans un fichier temporaire\n",
" with tempfile.NamedTemporaryFile(suffix=\".py\", delete=False, mode='w', encoding='utf-8') as tmp_file:\n",
" tmp_filename = tmp_file.name\n",
" tmp_file.write(code)\n",
" \n",
" try:\n",
" # Exécution du code dans le répertoire courant\n",
" result = subprocess.run(\n",
" [\"python\", tmp_filename],\n",
" capture_output=True,\n",
" text=True,\n",
" check=False # On met check=False pour capturer l'erreur sans lever l'exception\n",
" )\n",
" stdout = result.stdout\n",
" stderr = result.stderr\n",
" finally:\n",
" os.remove(tmp_filename)\n",
" \n",
" return stdout, stderr\n",
"\n",
"# --------------------------------------------------------------------\n",
"# 4) Fonction pour interpréter les résultats en streaming Markdown\n",
"# + Possibilité de joindre une image (ou plusieurs) depuis generated_outputs\n",
"# --------------------------------------------------------------------\n",
"def interpret_results_streaming(proof_text, generated_code, stdout, stderr):\n",
" \"\"\"\n",
" Envoie le code (generated_code) et les résultats de l'exécution (stdout, stderr) à l'API pour une interprétation textuelle\n",
" au regard de la preuve fournie, en mode streaming.\n",
"\n",
" Cette fois, on va automatiquement scanner le dossier 'generated_outputs'\n",
" pour chercher tous les .png. On les insère un par un dans le message.\n",
" \"\"\"\n",
" system_msg = (\n",
" \"Tu es un assistant IA spécialisé en mathématiques et en interprétation de résultats de simulation. \"\n",
" \"On te fournit la preuve initiale et le code d'une simulation ainsi que ses retours (generated_code, stdout, stderr). \"\n",
" \"Donne une analyse de la cohérence entre la preuve, le code et les résultats, \"\n",
" \"et retourne ta réponse au format Markdown.\"\n",
" )\n",
"\n",
" # On construit un 'content' qui est un tableau \n",
" # (selon l'exemple de code que vous avez fourni).\n",
" user_content = [\n",
" {\n",
" \"type\": \"text\",\n",
" \"text\": (\n",
" f\"Preuve initiale :\\n{proof_text}\\n\\n\"\n",
" f\"Code de la simulation :\\n{generated_code}\\n\\n\"\n",
" f\"Résultats (stdout) :\\n{stdout}\\n\\n\"\n",
" f\"Erreurs éventuelles (stderr) :\\n{stderr}\\n\\n\"\n",
" \"Merci d'interpréter ces résultats et de conclure sur la preuve. \"\n",
" \"Formule ta réponse de manière structurée en Markdown.\\n\"\n",
" )\n",
" }\n",
" ]\n",
"\n",
" # On cherche toutes les images PNG qui auraient pu être générées \n",
" # dans le dossier \"generated_outputs\"\n",
" output_dir = \"generated_outputs\"\n",
" png_files = glob.glob(os.path.join(output_dir, \"*.png\"))\n",
"\n",
" # Pour chacune, on l'encode en base64 et on l'ajoute\n",
" for png_path in png_files:\n",
" encoded_img = encode_image(png_path)\n",
" # On ajoute un bloc \"image_url\"\n",
" user_content.append({\n",
" \"type\": \"image_url\",\n",
" \"image_url\": {\"url\": f\"data:image/png;base64,{encoded_img}\"}\n",
" })\n",
"\n",
" # Appel en mode streaming\n",
" response_stream = openai.chat.completions.create(\n",
" model=MODEL_NAME,\n",
" messages=[\n",
" {\"role\": \"system\", \"content\": system_msg},\n",
" {\"role\": \"user\", \"content\": user_content}\n",
" ],\n",
" temperature=0.2,\n",
" stream=True\n",
" )\n",
" \n",
" # On débute un bloc Markdown\n",
" response = \"\"\n",
" display_handle = display(Markdown(\"\"), display_id=True)\n",
" for chunk in response_stream:\n",
" response += chunk.choices[0].delta.content or ''\n",
" response = response.replace(\"```\",\"\").replace(\"markdown\", \"\")\n",
" update_display(Markdown(response), display_id=display_handle.display_id)\n",
"\n",
"# --------------------------------------------------------------------\n",
"# 5) Fonction principale (à appeler directement dans le notebook)\n",
"# --------------------------------------------------------------------\n",
"def main(proof_text: str):\n",
" print(\"=== Génération du code Python pour tester la preuve... ===\")\n",
" test_code = generate_test_code(proof_text)\n",
"\n",
" # --- Nettoyage des backticks Markdown ---\n",
" lines = test_code.splitlines()\n",
" cleaned_lines = []\n",
" for line in lines:\n",
" if line.strip().startswith(\"```\"):\n",
" continue\n",
" cleaned_lines.append(line)\n",
" test_code = \"\\n\".join(cleaned_lines).strip()\n",
" # ----------------------------------------\n",
"\n",
" print(\"\\n=== Code généré (nettoyé) : ===\")\n",
" print(test_code)\n",
"\n",
" print(\"\\n=== Exécution du code généré... ===\")\n",
" stdout, stderr = run_generated_code(test_code)\n",
"\n",
" print(\"\\n=== Sortie standard (stdout) : ===\")\n",
" print(stdout)\n",
" if stderr.strip():\n",
" print(\"\\n=== Erreurs (stderr) : ===\")\n",
" print(stderr)\n",
"\n",
" print(\"\\n=== Interprétation des résultats (streaming en Markdown) ===\")\n",
" interpret_results_streaming(proof_text, test_code, stdout, stderr)\n",
" \n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1120131d-48f1-4fdc-8950-70312b8228df",
"metadata": {},
"outputs": [],
"source": [
"mon_texte_de_preuve = \"\"\"\n",
"Ci-dessous, je propose une ébauche (relativement détaillée) d’une approche formelle pour construire un objet « cosmohedron-like » et démontrer (ou du moins argumenter rigoureusement) qu’il possède des propriétés fractales dans une limite bien définie. Attention : dans l’état actuel des recherches, la littérature ne fournit pas (à ma connaissance) de démonstration largement reconnue établissant qu’un « vrai » cosmohedron est strictement fractal. Ce que je vais donc présenter est un modèle mathématisé inspiré des idées de cosmohedra et de leur possible fractalité, en détaillant :\n",
"La définition axiomatique (ou construction) d’une famille {Cn}n∈N\\{C_n\\}_{n \\in \\mathbb{N}}{Cn​}n∈N​ d’objets géométriques (polytopes) qui tendent vers une limite.\n",
"Les propriétés de self-similarité ou d’auto-similarité approchée qui sont au cœur d’une structure fractale.\n",
"Une preuve (ou un argument) de non-invariance d’échelle integer-valued (c’est-à-dire que la dimension n’est pas un entier fixe) en s’appuyant sur une analyse de la « taille » de l’objet à différentes échelles.\n",
"Une conclusion sur la (quasi-)fractalité ou la fractalité effective de la limite de {Cn}\\{C_n\\}{Cn​}.\n",
"\n",
"1. Définition d’une famille de polytopes « cosmohedron-like »\n",
"1.1. Construction combinatoire\n",
"Les cosmohedra (au sens de la littérature actuelle) se définissent via des données combinatoires et cinématiques (angles, énergie, moment, variables conformes, etc.) associées à la fonction d’onde cosmologique. Pour formaliser, on peut s’inspirer d’une définition axiomatique :\n",
"On part d’un polygone (ou polytope) de base Π0\\Pi_0Π0​ en dimension ddd (avec d≥2d \\ge 2d≥2, souvent la dimension 2 sert d’analogue).\n",
"On définit une liste de coupes ou « partitions » (généralisant la notion de triangulation) qui correspond, dans le contexte de la fonction d’onde cosmologique, aux factorisations possibles en sous-problèmes (un parallèle direct avec les associaèdres pour les amplitudes).\n",
"À chaque coupure ou partition, on associe des variables α,β,…\\alpha, \\beta,\\dotsα,β,… (analogues à des Mandelstam variables ou à des invariants kinematiques).\n",
"Le cosmohedron en tant que polytope est l’intersection d’un certain nombre d’inégalités linéaires (et éventuellement quadratiques, selon les modèles) provenant de ces variables : { x∈RN:Mi(x)≥0 ∀i∈I}, \\big\\{\\, x\\in \\mathbb{R}^N : M_i(x)\\ge 0 \\ \\forall i \\in I\\big\\},{x∈RN:Mi​(x)≥0 ∀i∈I}, où chaque MiM_iMi​ capture une condition « physique » ou « combinatoire » de consistance.\n",
"Pour exhiber la structure fractale, on définit une famille {Cn}\\{C_n\\}{Cn​} (analogues aux cosmohedra de plus en plus « raffinés ») via un processus itératif :\n",
"Initialisation : C1C_1C1​ est un polytope de dimension ddd (par exemple, un associaèdre standard ou un polygone en 2D).\n",
"Itération : Pour passer de CnC_nCn​ à Cn+1C_{n+1}Cn+1​, on effectue :\n",
"L’introduction de nouvelles variables cinématiques (exemple : scission supplémentaire de l’énergie dans un canal de Feynman),\n",
"L’ajout de nouvelles inégalités linéaires (ou combinatoires) qui forcent la convexité dans un espace plus grand,\n",
"Le tout aboutit à un polytope Cn+1⊂RNn+1C_{n+1}\\subset \\mathbb{R}^{N_{n+1}}Cn+1​⊂RNn+1​.\n",
"De cette manière, dim⁡(Cn)\\dim(C_n)dim(Cn​) augmente (ou du moins, l’espace ambiant RNn\\mathbb{R}^{N_n}RNn​ grandit), tandis que la projection de Cn+1C_{n+1}Cn+1​ sur certains sous-espaces ressemble de plus en plus à plusieurs copies (déformées) du polytope CnC_nCn​. C’est cette “auto-similarité” (même si souvent approchée et non exacte) qui peut nous donner la clé d’une structure fractale.\n",
"\n",
"1.2. Hypothèse d’auto-similarité asymptotique\n",
"On formalise l’idée que « chaque nouveau polytope Cn+1C_{n+1}Cn+1​ contient plusieurs copies réduites de CnC_nCn​ ». Par exemple, on peut dire qu’il existe un nombre k≥2k\\ge 2k≥2 et un facteur d’échelle ρ∈(0,1)\\rho\\in (0,1)ρ∈(0,1) tels que, pour un grand nnn,\n",
"Cn+1≈⋃i=1k(Φi(Cn)), C_{n+1} \\approx \\bigcup_{i=1}^k \\big( \\Phi_i(C_n)\\big),Cn+1​≈i=1⋃k​(Φi​(Cn​)),\n",
"où Φi\\Phi_iΦi​ sont des transformations affines contractantes (i.e. ∥Φi(x)−Φi(y)∥≤ρ∥x−y∥\\|\\Phi_i(x) - \\Phi_i(y)\\|\\le \\rho\\|x-y\\|∥Φi​(x)−Φi​(y)∥≤ρ∥x−y∥).\n",
"Dans la littérature fractale, quand on a une famille d’applications contractantes Φ1,…,Φk\\Phi_1,\\dots,\\Phi_kΦ1​,…,Φk​, il existe un ensemble (dit attracteur fractal) F⊂RmF\\subset \\mathbb{R}^mF⊂Rm tel que\n",
"F=⋃i=1kΦi(F). F = \\bigcup_{i=1}^k \\Phi_i(F).F=i=1⋃k​Φi​(F).\n",
"Si l’on parvient à montrer que {Cn}\\{C_n\\}{Cn​} converge (dans une topologie appropriée) vers un tel ensemble FFF, et que FFF n’est ni purement de dimension topologique dim⁡top∈N\\dim_{\\text{top}}\\in \\mathbb{N}dimtop​∈N ni trop mince (comme un ensemble de mesure zéro trop trivial), alors on peut conclure que FFF est fractal. On se servirait, par exemple, des résultats classiques de théorie des IFS (Iterated Function Systems, Barnsley et Hutchinson, 1981-1982).\n",
"\n",
"2. Critères de fractalité et preuve formelle\n",
"Pour qu’un sous-ensemble F⊂RmF \\subset \\mathbb{R}^mF⊂Rm soit considéré comme « fractal », une définition classique (à la Falconer, ou à la Mandelbrot) repose sur :\n",
"Dimension de Hausdorff dim⁡H(F)\\dim_{\\mathcal{H}}(F)dimH​(F) non entière ou strictement plus grande que sa dimension topologique.\n",
"Un certain degré de self-similarité (exacte ou statistique).\n",
"La dimension de Hausdorff peut être calculée ou estimée via la méthode de Hutchinson :\n",
"Si FFF est l’attracteur d’un système d’applications contractantes {Φi}i=1k\\{\\Phi_i\\}_{i=1}^k{Φi​}i=1k​ avec un facteur d’échelle ρ<1\\rho < 1ρ<1 identique (ou ρi\\rho_iρi​ variables mais bornées) et certaines conditions de non-recouvrement trop fort (condition OSC — Open Set Condition), alors la dimension de Hausdorff dim⁡H(F)\\dim_{\\mathcal{H}}(F)dimH​(F) est la solution unique de :\n",
"∑i=1kρis  =  1, \\sum_{i=1}^k \\rho_i^s \\;=\\; 1,i=1∑k​ρis​=1,\n",
"où ρi\\rho_iρi​ est le plus grand facteur de contraction de Φi\\Phi_iΦi​. Généralement, la solution sss n’est pas un entier, d’où le caractère fractal.\n",
"\n",
"2.1. Argument de la preuve : suite de polytopes CnC_nCn​ convergente\n",
"Énoncé : Supposons que la suite (Cn)(C_n)(Cn​) de polytopes (chacun potentiellement en dimension différente, ou projetée dans une dimension ≤m\\le m≤m) soit telle que, pour des constantes ρ<1\\rho<1ρ<1 et un entier k≥2k\\ge2k≥2, on ait :\n",
"Cn+1⊂⋃i=1kΦn,i(Cn), C_{n+1} \\subset \\bigcup_{i=1}^k \\Phi_{n,i}(C_n),Cn+1​⊂i=1⋃k​Φn,i​(Cn​),\n",
"avec Φn,i\\Phi_{n,i}Φn,i​ une application (au moins) contractante, et que le chevauchement entre les Φn,i(Cn)\\Phi_{n,i}(C_n)Φn,i​(Cn​) reste contrôlé (afin de satisfaire une version “dynamique” de l’Open Set Condition).\n",
"Extraction d’un système contractant :\n",
"Si, pour chaque nnn, les Φn,i\\Phi_{n,i}Φn,i​ sont εn\\varepsilon_nεn​-proches d’applications Φi\\Phi_iΦi​ (indépendantes de nnn) et εn→0\\varepsilon_n \\to 0εn​→0, alors dans la limite n→∞n\\to\\inftyn→∞, on obtient un système d’applications {Φi}i=1k\\{\\Phi_i\\}_{i=1}^k{Φi​}i=1k​ fixes et contractantes.\n",
"Soit FFF l’attracteur fractal de ce système (au sens usuel de la théorie IFS). La suite {Cn}\\{C_n\\}{Cn​} peut alors être montrée convergente (par exemple, pour la distance de Hausdorff sur les compacts) vers l’ensemble FFF.\n",
"Résultat : dim⁡H(F)=s\\dim_{\\mathcal{H}}(F) = sdimH​(F)=s, où sss est la solution de l’équation de Hutchinson :\n",
"∑i=1kρis=1, \\sum_{i=1}^k \\rho_i^s = 1,i=1∑k​ρis​=1,\n",
"(supposé non entier). Alors FFF est fractal, et CnC_nCn​ “devient” fractal dans la limite.\n",
"Conséquence : On a donc formellement exhibé un objet (la limite) qui n’a plus de dimension Euclidienne standard, mais une dimension de Hausdorff non entière, possédant un motif de répétition (self-similarité). C’est précisément ce que l’on entend par « fractal » dans un sens rigoureux.\n",
"\n",
"2.2. Hypothèses nécessaires\n",
"(H1) Contractions : Les Φn,i\\Phi_{n,i}Φn,i​ doivent réellement contracter les distances (ex. affinité avec un facteur ρ<1\\rho<1ρ<1).\n",
"(H2) Contrôle de recouvrement : Il ne doit pas y avoir un trop grand recouvrement ou une accumulation pathologique (sinon la dimension de Hausdorff peut diverger ou se réduire à un objet trop simple).\n",
"(H3) Approximation stable : On suppose que la suite Φn,i\\Phi_{n,i}Φn,i​ converge (au moins localement) vers {Φi}i=1k\\{\\Phi_i\\}_{i=1}^k{Φi​}i=1k​, ce qui permet de “geler” la dynamique dans la limite.\n",
"Ces hypothèses sont, dans la pratique, difficiles à vérifier précisément pour les véritables cosmohedra ; elles sont plus simples à démontrer pour un modèle qui capture les mêmes règles combinatoires et dont la géométrie (les inégalités) est choisie pour permettre ces propriétés.\n",
"\n",
"3. Application à un « modèle cosmohedron fractal »\n",
"Pour aller du formalisme théorique ci-dessus à un exemple concret, on peut définir explicitement :\n",
"Un ensemble de variables (t1,…,tn)(t_1, \\dots, t_n)(t1​,…,tn​) modélisant les différents canaux d’énergie/moment (analogie aux coupes Feynman).\n",
"Un polytope Cn⊂RnC_n\\subset \\mathbb{R}^nCn​⊂Rn défini par des inégalités du type 0≤t1≤t2≤⋯≤tn≤10 \\le t_1 \\le t_2 \\le \\dots \\le t_n \\le 10≤t1​≤t2​≤⋯≤tn​≤1 et des contraintes supplémentaires (ti+ti+1≤α ti−1+β)(t_i + t_{i+1} \\le \\alpha\\,t_{i-1} + \\beta)(ti​+ti+1​≤αti−1​+β), etc.\n",
"Règles de subdivision : pour construire Cn+1C_{n+1}Cn+1​ à partir de CnC_nCn​, on rajoute des variables tn+1,…,tn+kt_{n+1}, \\dots, t_{n+k}tn+1​,…,tn+k​ et des inégalités analogues, de sorte que la projection sur {t1,…,tn}\\{t_1,\\dots, t_n\\}{t1​,…,tn​} se décompose en “copies” échelonnées de CnC_nCn​.\n",
"Exemple schématique :\n",
"Cn+1  =  ⋂i=1n+1{(t1,…,tn+1):ti≥0,  ∑j=1n+1tj=1,  … }, C_{n+1} \\;=\\; \\bigcap_{i=1}^{n+1} \\bigl\\{ (t_1,\\dots,t_{n+1}) : t_{i} \\ge 0,\\; \\sum_{j=1}^{n+1} t_j = 1,\\; \\dots \\bigr\\},Cn+1​=i=1⋂n+1​{(t1​,…,tn+1​):ti​≥0,j=1∑n+1​tj​=1,…},\n",
"avec certaines conditions linéaires (ou affines) introduisant une auto-similarité. On peut prouver que, si on choisit bien les coefficients, alors on obtient une suite {Cn}\\{C_n\\}{Cn​} satisfaisant les hypothèses (H1), (H2) et (H3).\n",
"La preuve de fractalité (au sens Hausdorff) s’articule alors sur la démonstration que la projection (ou section) dans un sous-espace de dimension 2 (ou plus) admet un recouvrement par Φn,i(Cn)\\Phi_{n,i}(C_n)Φn,i​(Cn​) avec un rapport d’échelle ρ<1\\rho<1ρ<1. Dès lors, la même théorie IFS s’applique et conclut qu’on obtient un attracteur fractal dans la limite.\n",
"\n",
"4. Conclusion et perspectives\n",
"Construction rigoureuse :\n",
"On a donné le schéma d’une construction formelle (suivant la logique de la théorie IFS) permettant de définir une suite de polytopes s’apparentant à une “version fractale” de cosmohedra. Les conditions de contraction et de non-recouvrement permettent l’utilisation des théorèmes standards de la géométrie fractale (Barnsley, Falconer, Hutchinson).\n",
"Preuve de la fractalité :\n",
"La démonstration repose sur la convergence vers un attracteur fractal  F\\,FF et l’évaluation de la dimension de Hausdorff par la formule ∑i=1kρis=1\\sum_{i=1}^k \\rho_i^s = 1∑i=1k​ρis​=1. On obtient ainsi, en général, un exposant sss non entier, prouvant le caractère fractal.\n",
"Lien avec la physique :\n",
"D’un point de vue strictement mathématique, l’existence ou non d’une fractalité dans le vrai “espace de configurations” d’un cosmohedron cosmologique reste conjecturale. Il faudrait prouver qu’en dimension (potentiellement) très élevée et avec des contraintes physiques (pôles d’amplitudes, invariances conformes, etc.), la construction reproduit les conditions de l’IFS.\n",
"Remarque finale :\n",
"Bien que cette approche donne un cadre théorique pour exhiber un objet fractal (et prouver rigoureusement ses propriétés de fractalité), son application exacte aux cosmohedra décrits dans la littérature de la fonction d’onde cosmologique exigerait des travaux d’adaptation. Néanmoins, c’est ainsi que l’on procéderait pour avoir un argument formel :\n",
"Montrer que la “croissance” des polytopes répond à un schéma de self-similarité (au moins asymptotique),\n",
"Démontrer, via un théorème standard d’attracteur IFS, que la dimension du lieu-limite est non entière,\n",
"Conclure que la structure est fractale.\n",
"En résumé, cette construction est rigoureuse si l’on respecte les hypothèses de contraction, de non-recouvrement et de convergence vers des transformations Φi\\Phi_iΦi​ fixes. Elle aboutit à une preuve formelle (dans le sens de la théorie des IFS) que le limite de la suite de polytopes est un ensemble fractal, et donc qu’il existe une structure fractale sous-jacente dans ce modèle “cosmohedron-like” à la limite n→∞n \\to \\inftyn→∞.\n",
"\n",
"\"\"\"\n",
"resultats_md = main(mon_texte_de_preuve)\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "44349990-50c7-4fb7-a62c-532d829c2bdc",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.11"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

197
week2/community-contributions/week2-exercise-translator.ipynb Normal file
View File

@@ -0,0 +1,197 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "7563a171",
"metadata": {},
"source": []
},
{
"cell_type": "markdown",
"id": "d006b2ea-9dfe-49c7-88a9-a5a0775185fd",
"metadata": {},
"source": [
"# Exercise - week 2: German translator\n",
"\n",
"This should include a Gradio UI, streaming, use of the system prompt to add expertise, and the ability to switch between models. Bonus points if you can demonstrate use of a tool!\n",
"\n",
"The assistant will transform your spoken English to text, then translate it German and speak it out. The image on the UI is just decoration. This exercise was created on MacOS, Python 3.13."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a07e7793-b8f5-44f4-aded-5562f633271a",
"metadata": {},
"outputs": [],
"source": [
"# Install first PortAudio, in MacOS\n",
"# brew install portaudio\n",
"\n",
"\n",
"!pip install openai speechrecognition pyaudio\n"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "dcae50aa",
"metadata": {},
"outputs": [],
"source": [
"# imports\n",
"\n",
"import os\n",
"import json\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import gradio as gr"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1796b554",
"metadata": {},
"outputs": [],
"source": [
"# Initialization\n",
"\n",
"load_dotenv()\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}\")\n",
"else:\n",
" print(\"OpenAI API Key not set\")\n",
" \n",
"MODEL = \"gpt-4o-mini\"\n",
"openai = OpenAI()"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "c5caad24",
"metadata": {},
"outputs": [],
"source": [
"system_message = \"\"\"You are a highly skilled language translator specializing in translating English text to German. \n",
"Your task is to accurately translate any English text provided by the user into German. \n",
"Ensure that the translations are grammatically correct and contextually appropriate. \n",
"If the user provides a phrase, sentence, or paragraph in English, respond with the equivalent translation in German.\"\"\" "
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "aca69563",
"metadata": {},
"outputs": [],
"source": [
"import speech_recognition as sr\n",
"from pydub import AudioSegment\n",
"from pydub.playback import play\n",
"import base64\n",
"from io import BytesIO\n",
"from PIL import Image\n",
"\n",
"\n",
"def recognize_speech(audio_file):\n",
" recognizer = sr.Recognizer()\n",
" with sr.AudioFile(audio_file) as source:\n",
" audio = recognizer.record(source)\n",
" try:\n",
" text = recognizer.recognize_google(audio)\n",
" return text\n",
" except sr.UnknownValueError:\n",
" return \"Google Speech Recognition could not understand audio\"\n",
" except sr.RequestError as e:\n",
" return f\"Could not request results from Google Speech Recognition service; {e}\"\n",
"\n",
"\n",
"def get_chatgpt_response(message):\n",
" response = openai.chat.completions.create(\n",
" model=MODEL,\n",
" messages = \n",
" [{\"role\": \"system\", \"content\": system_message},\n",
" {\"role\": \"user\", \"content\": message}],\n",
" max_tokens=150\n",
" )\n",
" return response.choices[0].message.content.strip()\n",
"\n",
"def process_audio(audio_file):\n",
" text = recognize_speech(audio_file)\n",
" if text:\n",
" response = get_chatgpt_response(text)\n",
" talker(response)\n",
" return response\n",
" return \"Could not recognize speech.\"\n",
"\n",
"def talker(message):\n",
" response = openai.audio.speech.create(\n",
" model=\"tts-1\",\n",
" voice=\"onyx\", # Also, try replacing onyx with alloy\n",
" input=message\n",
" )\n",
" \n",
" audio_stream = BytesIO(response.content)\n",
" audio = AudioSegment.from_file(audio_stream, format=\"mp3\")\n",
" play(audio)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f1118141",
"metadata": {},
"outputs": [],
"source": [
"# Create Gradio interface\n",
"\n",
"# some image decoration to UI, just a static picture\n",
"image_path =\"week2-exercise-translator-berlin.webp\"\n",
"\n",
"with gr.Blocks() as ui:\n",
" gr.Interface(\n",
" fn=process_audio,\n",
" inputs=gr.Audio(type=\"filepath\", label=\"Speak English. German translation in a moment:\"),\n",
" outputs=\"text\",\n",
" live=True, \n",
" )\n",
" gr.Image(value=image_path, label=\"Das ist Berlin\")\n",
" \n",
"ui.launch(inbrowser=True)\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c1284da5",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "venv313",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.13.2"
}
},
"nbformat": 4,
"nbformat_minor": 5
}

614
week2/community-contributions/week2_exercise_translated_chatbot.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

22
week2/day1.ipynb
View File

@@ -300,6 +300,7 @@
"source": [
"# Claude 3.5 Sonnet again\n",
"# Now let's add in streaming back results\n",
"# If the streaming looks strange, then please see the note below this cell!\n",
"\n",
"result = claude.messages.stream(\n",
" model=\"claude-3-5-sonnet-latest\",\n",
@@ -316,6 +317,27 @@
" print(text, end=\"\", flush=True)"
]
},
{
"cell_type": "markdown",
"id": "dd1e17bc-cd46-4c23-b639-0c7b748e6c5a",
"metadata": {},
"source": [
"## A rare problem with Claude streaming on some Windows boxes\n",
"\n",
"2 students have noticed a strange thing happening with Claude's streaming into Jupyter Lab's output -- it sometimes seems to swallow up parts of the response.\n",
"\n",
"To fix this, replace the code:\n",
"\n",
"`print(text, end=\"\", flush=True)`\n",
"\n",
"with this:\n",
"\n",
"`clean_text = text.replace(\"\\n\", \" \").replace(\"\\r\", \" \")` \n",
"`print(clean_text, end=\"\", flush=True)`\n",
"\n",
"And it should work fine!"
]
},
{
"cell_type": "code",
"execution_count": null,

4
week2/day2.ipynb
View File

@@ -53,7 +53,7 @@
"# Load environment variables in a file called .env\n",
"# Print the key prefixes to help with any debugging\n",
"\n",
"load_dotenv()\n",
"load_dotenv(override=True)\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",
@@ -130,6 +130,8 @@
"metadata": {},
"outputs": [],
"source": [
"# This can reveal the \"training cut off\", or the most recent date in the training data\n",
"\n",
"message_gpt(\"What is today's date?\")"
]
},

2
week2/day3.ipynb
View File

@@ -33,7 +33,7 @@
"# Load environment variables in a file called .env\n",
"# Print the key prefixes to help with any debugging\n",
"\n",
"load_dotenv()\n",
"load_dotenv(override=True)\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"google_api_key = os.getenv('GOOGLE_API_KEY')\n",

2
week2/day4.ipynb
View File

@@ -35,7 +35,7 @@
"source": [
"# Initialization\n",
"\n",
"load_dotenv()\n",
"load_dotenv(override=True)\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",

2
week2/day5.ipynb
View File

@@ -35,7 +35,7 @@
"source": [
"# Initialization\n",
"\n",
"load_dotenv()\n",
"load_dotenv(override=True)\n",
"\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"if openai_api_key:\n",