Merge pull request #816 from davebelliveau/week2_day1_coffee_talk
Add week 2, day 1, N way conversation solution
This commit is contained in:
Ed Donner
GitHub
@@ -0,0 +1,283 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"id": "88f67391",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"### N Way Conversation - Coffee Talk \n",
|
||||
"\n",
|
||||
"This example simulates an N-way conversation between the characters of the Saturday Night Live skit Coffee Talk.\n",
|
||||
"\n",
|
||||
"The character information is retrieved from a model and each character is handled by its own model selected at random from a list of available models. Only the number of characters, number of rounds, and available models are configured.\n",
|
||||
"\n",
|
||||
"The example can use OpenRouter, OpenAI, or Ollama, in that order. "
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "a1eeb029",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# Setup ...\n",
|
||||
"\n",
|
||||
"# The number of characters (models) conversing\n",
|
||||
"NBR_CHARACTERS=4\n",
|
||||
"\n",
|
||||
"# The number of rounds of conversation\n",
|
||||
"NBR_ROUNDS=4\n",
|
||||
"\n",
|
||||
"# Available OpenRouter models. The base model is used to select characters and the topic. Other models are used for the conversation\n",
|
||||
"OPENROUTER_MODELS=\"openai/gpt-4.1-mini, anthropic/claude-3.5-haiku, google/gemini-2.5-flash\"\n",
|
||||
"OPENROUTER_BASE=\"openai/gpt-5\"\n",
|
||||
"\n",
|
||||
"# Available OpenAI models\n",
|
||||
"OPENAI_MODELS=\"gpt-4.1, gpt-4.1-mini, gpt-5-nano\"\n",
|
||||
"OPENAI_BASE=\"gpt-5\"\n",
|
||||
"\n",
|
||||
"# Available Ollama models. Note that these must be pre-fetched or errors will occur (and won't be handled)\n",
|
||||
"OLLAMA_MODELS=\"gpt-oss, gemma3, llama3.2\"\n",
|
||||
"OLLAMA_BASE=\"gpt-oss\"\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "68022fbc",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# imports\n",
|
||||
"import os\n",
|
||||
"import json\n",
|
||||
"from dotenv import load_dotenv\n",
|
||||
"from IPython.display import Markdown, display, update_display\n",
|
||||
"from openai import OpenAI"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "73460c5e",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# Setup the LLM client and models. OpenRouter has priority if available, then OpenAI, then Ollama.\n",
|
||||
"\n",
|
||||
"load_dotenv(override=True)\n",
|
||||
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
|
||||
"openrouter_api_key = os.getenv('OPENROUTER_API_KEY')\n",
|
||||
"\n",
|
||||
"if openrouter_api_key:\n",
|
||||
" print(f\"OpenRouter API Key exists and begins {openrouter_api_key[:3]}, using OpenRouter.\")\n",
|
||||
" available_models=OPENROUTER_MODELS\n",
|
||||
" base_model=OPENROUTER_BASE\n",
|
||||
" client = OpenAI(base_url=\"https://openrouter.ai/api/v1\", api_key=openrouter_api_key)\n",
|
||||
"elif openai_api_key:\n",
|
||||
" print(f\"OpenAI API Key exists and begins {openai_api_key[:8]}, using OpenAI.\")\n",
|
||||
" available_models=OPENAI_MODELS\n",
|
||||
" base_model=OPENAI_BASE\n",
|
||||
" client = OpenAI()\n",
|
||||
"else:\n",
|
||||
" print(\"OpenAI API Key not set, using Ollama.\")\n",
|
||||
" available_models=OLLAMA_MODELS\n",
|
||||
" base_model=OLLAMA_BASE\n",
|
||||
" client = OpenAI(api_key=\"ollama\", base_url=\"http://localhost:11434/v1\")"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "b1a7004d",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# Get the characters from the base model\n",
|
||||
"system_prompt = \"\"\"\n",
|
||||
"You will be asked to return information about characters in the SNL skit Coffee Talk\n",
|
||||
"You should return the information as a JSON response with the following format:\n",
|
||||
"{\n",
|
||||
" { \"name\" : \"Linda\", \"persona\", \"....\", \"model\" : \"model-name\" },\n",
|
||||
" { \"name\" : \"Paul\", \"persona\", \"....\", \"model\" : \"model-name\" }\n",
|
||||
"}\n",
|
||||
"\n",
|
||||
"\"\"\"\n",
|
||||
"\n",
|
||||
"user_prompt = f\"\"\"\n",
|
||||
"Create a list of the many characters from the SNL skit Coffee Talk, and return {NBR_CHARACTERS} total characters.\n",
|
||||
"Always return Linda Richmond as the first character.\n",
|
||||
"Return one caller.\n",
|
||||
"Select the remaining characters at random from the list of all characters. \n",
|
||||
"For the model value, return a random model name from this list: {available_models}.\n",
|
||||
"\"\"\"\n",
|
||||
"\n",
|
||||
"response = client.chat.completions.create(\n",
|
||||
" model=base_model,\n",
|
||||
" messages=[\n",
|
||||
" {\"role\": \"system\", \"content\": system_prompt},\n",
|
||||
" {\"role\": \"user\", \"content\": user_prompt}\n",
|
||||
" ],\n",
|
||||
" response_format={\"type\": \"json_object\"}\n",
|
||||
" )\n",
|
||||
"result = response.choices[0].message.content\n",
|
||||
"characters = json.loads(result)\n",
|
||||
"\n",
|
||||
"print(json.dumps(characters, indent=2))\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "21a73805",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# Generate system prompts for each character, which includes their name, persona, the other guests, and how they should respond.\n",
|
||||
"\n",
|
||||
"guests = \"The guests on todays show are \"\n",
|
||||
"guest_names = [character['name'] for character in characters[\"characters\"]]\n",
|
||||
"guests += \", \".join(guest_names)\n",
|
||||
"\n",
|
||||
"prompt = \"\"\n",
|
||||
"for character in characters[\"characters\"]:\n",
|
||||
" prompt = f\"You are {character['name']} a character on the SNL skit Coffee Talk.\"\n",
|
||||
" prompt += f\" Your personality is : {character['persona']} \"\n",
|
||||
" prompt += \" \" + guests + \".\"\n",
|
||||
" prompt += \" Keep responses brief and in character.\"\n",
|
||||
" prompt += \" In the conversation history, each response is prefixed with the character's name to identify the respondent.\"\n",
|
||||
" prompt += \" Your response should not include your character name as a prefix.\"\n",
|
||||
"\n",
|
||||
" character[\"system_prompt\"] = prompt\n",
|
||||
"\n",
|
||||
"print(json.dumps(characters, indent=2))\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "656131a1",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# Get the topic\n",
|
||||
"user_prompt=\"\"\"\n",
|
||||
"In the SNL skit Coffee Talk, the host Linda Richmond proposes topics in the form \"X Y is neither X, nor Y - discuss\".\n",
|
||||
"Create a list of the many topics proposed on the show, and select one at random and return it.\n",
|
||||
"Return only the selected topic without any formatting.\n",
|
||||
"\"\"\"\n",
|
||||
"\n",
|
||||
"response = client.chat.completions.create(\n",
|
||||
" model=base_model,\n",
|
||||
" messages=[\n",
|
||||
" {\"role\": \"user\", \"content\": user_prompt}\n",
|
||||
" ],\n",
|
||||
" )\n",
|
||||
"topic = response.choices[0].message.content\n",
|
||||
"\n",
|
||||
"print(topic)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "6e137753",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"def get_character_response(character,history):\n",
|
||||
" user_prompt = f\"\"\"\n",
|
||||
" The conversation so far is as follows:\n",
|
||||
" {history}\n",
|
||||
" What is your response? \n",
|
||||
" \"\"\"\n",
|
||||
" \n",
|
||||
" response = client.chat.completions.create(\n",
|
||||
" model=character[\"model\"],\n",
|
||||
" messages=[\n",
|
||||
" {\"role\": \"system\", \"content\": character[\"system_prompt\"]},\n",
|
||||
" {\"role\": \"user\", \"content\": user_prompt}\n",
|
||||
" ]\n",
|
||||
" )\n",
|
||||
" return response.choices[0].message.content\n",
|
||||
" "
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"id": "23fb446f",
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# Start the show!\n",
|
||||
"\n",
|
||||
"history = \"\"\n",
|
||||
"history += \"Welcome to Coffee Talk, I am your host Linda Richmond. Today's guests are:\\n\"\n",
|
||||
"\n",
|
||||
"for character in characters[\"characters\"][1:]:\n",
|
||||
" history += f\" - {character['name']}\\n\"\n",
|
||||
"\n",
|
||||
"history += f\"\\nI'll give you a topic: {topic}\\n\"\n",
|
||||
"\n",
|
||||
"display(Markdown(\"---\"))\n",
|
||||
"display(Markdown(history))\n",
|
||||
"display(Markdown(\"---\"))\n",
|
||||
"\n",
|
||||
"# Other guests respond (first round)\n",
|
||||
"for character in characters[\"characters\"][1:]:\n",
|
||||
" response = get_character_response(character,history)\n",
|
||||
" display(Markdown(f\"**{character['name']}({character['model']}):** {response}\")) \n",
|
||||
" history += f\"\\n{character['name']}: {response}\"\n",
|
||||
"\n",
|
||||
"# Continue conversation for remaining rounds (all characters including Linda)\n",
|
||||
"for round in range(1, NBR_ROUNDS):\n",
|
||||
" for character in characters[\"characters\"]:\n",
|
||||
" response = get_character_response(character,history)\n",
|
||||
" display(Markdown(f\"**{character['name']}({character['model']}):** {response}\")) \n",
|
||||
" history += f\"\\n{character['name']}: {response}\"\n",
|
||||
"\n",
|
||||
"# Wrap it up\n",
|
||||
"user_prompt=f\"\"\"\n",
|
||||
"It's time to wrap up the show. Here's the whole conversation:\\n\n",
|
||||
"{history}\n",
|
||||
"Wrap up the show, as only you can.\n",
|
||||
"\"\"\"\n",
|
||||
"\n",
|
||||
"linda = characters[\"characters\"][0]\n",
|
||||
"response = client.chat.completions.create(\n",
|
||||
" model=linda[\"model\"],\n",
|
||||
" messages=[\n",
|
||||
" {\"role\": \"system\", \"content\": linda[\"system_prompt\"]},\n",
|
||||
" {\"role\": \"user\", \"content\": user_prompt}\n",
|
||||
" ]\n",
|
||||
" )\n",
|
||||
"\n",
|
||||
"display(Markdown(\"---\"))\n",
|
||||
"display(Markdown(response.choices[0].message.content)) \n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"kernelspec": {
|
||||
"display_name": "llm-engineering",
|
||||
"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": "3.12.11"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 5
|
||||
}
|
||||
Reference in New Issue
Block a user