Spaces:
Sleeping
Sleeping
File size: 13,381 Bytes
5fdb69e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 |
{
"cells": [
{
"cell_type": "markdown",
"id": "5df0164c-1980-4fd7-94e4-a71b485a41fd",
"metadata": {},
"source": [
"# Week 2 Day 1 - Conversation between three AI's\n",
"\n",
"This notebook defines three classes (`ThreeWayChat`, `Participant` and `Model`) that implement a 3-way conversation between different AI's. \n",
"\n",
"At the bottom there is an example conversation between a Claude model and two GPT models.\n",
"\n",
"The implementation works with models available via the `openai` and `anthropic` libraries."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8b466547-809a-4b81-bfd7-ce9a1ac4bb2b",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import logging\n",
"import re\n",
"\n",
"from dotenv import load_dotenv\n",
"from openai import OpenAI\n",
"import anthropic"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "acaff46f-e43e-4527-a404-a5b3ae830e51",
"metadata": {},
"outputs": [],
"source": [
"logging.basicConfig(\n",
" level=logging.WARNING,\n",
" format=\"%(levelname)s:%(name)s:%(funcName)s:%(message)s\"\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "aca57918-0271-4574-918b-2808f51698d1",
"metadata": {},
"outputs": [],
"source": [
"# check if API keys are in .env\n",
"load_dotenv(override=True)\n",
"openai_api_key = os.getenv('OPENAI_API_KEY')\n",
"anthropic_api_key = os.getenv('ANTHROPIC_API_KEY')\n",
"\n",
"assert openai_api_key, \"OpenAI API key is missing\"\n",
"assert anthropic_api_key, \"Anthropic API key is missing\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "25c37440-8692-4a8d-95e6-998691b4acf6",
"metadata": {},
"outputs": [],
"source": [
"class Model:\n",
" \"\"\"One class for different API's.\n",
" \n",
" This implementation allows the use of the OpenAI and Anthropic API. Other endpoints,\n",
" such as Ollama, can be used as well, as long as they are used via the OpenAI\n",
" Python library.\n",
" \n",
" \"\"\"\n",
" def __init__(self, api=None, model_name=\"mock\"):\n",
" \"\"\"\n",
" Args:\n",
" api: Can be an OpenAI or anthropic.Anthropic object or None to make a mock run.\n",
" model_name (str): Identifies the model used via the API.\n",
"\n",
" \"\"\"\n",
" self.api = api\n",
" self.name = model_name\n",
" if type(self.api) not in {OpenAI, anthropic.Anthropic} and self.name not in {\"mock\", \"\"}:\n",
" logging.warning(f\"Unknown API '{self.api}'. Using mock.\")\n",
"\n",
" def complete(self, messages, system=\"\"):\n",
" \"\"\"Make API call.\"\"\"\n",
" completion = \"\"\n",
" if isinstance(self.api, OpenAI):\n",
" completion = self.api.chat.completions.create(\n",
" model=self.name,\n",
" messages=[{\"role\": \"system\", \"content\": system}] + messages,\n",
" max_tokens=300\n",
" )\n",
" completion = completion.choices[0].message.content\n",
"\n",
" elif isinstance(self.api, anthropic.Anthropic):\n",
" completion = self.api.messages.create(\n",
" model=self.name,\n",
" system=system,\n",
" messages=messages,\n",
" max_tokens=300\n",
" )\n",
" completion = completion.content[0].text\n",
" \n",
" else:\n",
" completion = \"Mock answer.\"\n",
"\n",
" return self.parse_answer(completion)\n",
"\n",
" def parse_answer(self, answer):\n",
" # Remove prefix 'Name:' from answer if present.\n",
" regex = r\"(?P<name>\\w+): (?P<content>.*)\"\n",
" match = re.match(regex, answer, re.DOTALL)\n",
" if match:\n",
" logging.info(f\"{self.name} generated {match.group('name')}\")\n",
" return match.group(\"content\")\n",
" return answer\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "462df0ba-36b5-4043-b0d0-a1d68edb968a",
"metadata": {},
"outputs": [],
"source": [
"class Participant:\n",
" \"\"\"Represents one participant in a conversation.\"\"\"\n",
" def __init__(self, name, model=Model(), system_prompt=\"\", initial_message=\"\"):\n",
" \"\"\"\n",
" Args:\n",
" model (Model): The model that is called to get participant's answer.\n",
" name (str): Used to assign answers to different participants. Is inserted in the\n",
" messages list, so the model knows who's spoken. Is also\n",
" displayed in the output.\n",
" system_prompt (str): The system prompt overgiven to the model backend.\n",
" initial_message (str): An optional conversation start.\n",
" \"\"\"\n",
" self.model = model\n",
" self.name = name\n",
" self.role = system_prompt\n",
" self.initial_msg = initial_message\n",
" self.messages = [] # keeps conversation history\n",
" self.last_msg = \"\"\n",
"\n",
" def speak(self):\n",
" if self.initial_msg:\n",
" self.last_msg = self.initial_msg\n",
" self.initial_msg = \"\"\n",
" else:\n",
" self.last_msg = self.model.complete(self.messages, self.role)\n",
" self.update_messages(role=\"assistant\", content=self.last_msg)\n",
" return self.last_msg\n",
"\n",
" def listen(self, message: str, speaker_name: str):\n",
" # Insert the speaker name, so the model can distinguish them\n",
" self.update_messages(role=\"user\", content=f\"{speaker_name}: {message}\")\n",
"\n",
" def update_messages(self, role, content):\n",
" self.messages.append({\"role\": role, \"content\": content})\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e838901f-9a50-4f6b-b30f-e78c27e86bd7",
"metadata": {},
"outputs": [],
"source": [
"class ThreeWayChat:\n",
" \"\"\"Make three Participants communicate.\"\"\"\n",
" def __init__(self, participants, n_turns=4):\n",
" \"\"\"\n",
" Args:\n",
" participants (tuple[Participant]): Three objects. The order determines the speaking order.\n",
" n_turns (int): Number of turns per participant, incl. Participant.initial_message.\n",
"\n",
" \"\"\"\n",
" self.n_turns = n_turns\n",
" self.p1, self.p2, self.p3 = participants\n",
" if len({bool(self.p1.initial_msg), bool(self.p2.initial_msg), bool(self.p3.initial_msg)}) != 1:\n",
" logging.warning(\"At least one Participant has gotten a value for initial_message while another hasn't.\")\n",
" if len({self.p1.name, self.p2.name, self.p3.name}) != 3:\n",
" raise ValueError(f\"Some Participants have the same name. \"\n",
" f\"Please use unique names.\"\n",
" f\"\\nNames you've given: {self.p1.name}, {self.p2.name} and {self.p3.name}. \")\n",
"\n",
" def start(self, n_turns=None):\n",
" \"\"\"Start a conversation with n_turns rounds.\n",
" \n",
" Args:\n",
" n_turns (int): If None, self.n_turns is used.\n",
"\n",
" \"\"\"\n",
" for i in range(n_turns or self.n_turns):\n",
" # Make each participant speak and display their answers\n",
" self.make_display_turn(self.p1, self.p2, self.p3)\n",
" self.make_display_turn(self.p2, self.p1, self.p3)\n",
" self.make_display_turn(self.p3, self.p2, self.p1)\n",
"\n",
" def make_display_turn(self, speaker, *listeners):\n",
" self.speaker_to_listeners(speaker, *listeners)\n",
" self.display_last_utterance(speaker)\n",
" \n",
" def speaker_to_listeners(self, speaker, *listeners):\n",
" \"\"\"Get answer from speaker and update conversation histories.\"\"\"\n",
" speaker_text = speaker.speak()\n",
" for listener in listeners:\n",
" listener.listen(speaker_text, speaker.name)\n",
"\n",
" def display_last_utterance(self, speaker):\n",
" print(\"{} ({}):\\n{}\\n\".format(\n",
" speaker.name.upper(), speaker.model.name, speaker.last_msg\n",
" ))\n",
"\n"
]
},
{
"cell_type": "markdown",
"id": "80294493-04ff-4bec-af88-c3fc11d21c54",
"metadata": {},
"source": [
"#### Example system prompts:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "997841b1-d547-472b-a298-a60be2f9b90f",
"metadata": {},
"outputs": [],
"source": [
"name1 = \"Austin\"\n",
"name2 = \"Jonas\"\n",
"name3 = \"Tim\"\n",
"\n",
"general_system = (\n",
" \"\\n\\nYou've entered a chatroom with two other participants. \"\n",
" 'Their names are \"{}\" and \"{}\". Your name is \"{}\".'\n",
" \"\\nGenerate a maximum of 100 words per turn.\"\n",
")\n",
"\n",
"system1 = (\n",
" \"You are very argumentative; \"\n",
" \"You always find something to discuss. \"\n",
" \"When someone says their opinion, you often disagree. \"\n",
" \"You enjoy swimming against the tide and mocking mainstream opinions.\"\n",
" + general_system.format(name3, name2, name1)\n",
")\n",
"\n",
"system2 = (\n",
" \"You have a very conservative and clear opinion on most things. \"\n",
" \"You feel safest in your familiar surroundings. You are very reluctant to try out new things. \"\n",
" \"In discourses you are stubborn and want to convince others from your gridlocked beliefs.\"\n",
" + general_system.format(name1, name3, name2)\n",
")\n",
"\n",
"system3 = (\n",
" \"You are very humorous and like to be ironic. Sometimes you tell silly jokes. \"\n",
" \"You like variation; If a discussion about a topic takes too long, you start a new topic.\"\n",
" + general_system.format(name1, name2, name3)\n",
")"
]
},
{
"cell_type": "markdown",
"id": "0f455bb6-c6a8-4f75-a003-4bfda8dcff8a",
"metadata": {},
"source": [
"#### Example with **Claude-3-Haiku** and *two instances* of **GPT-4o-mini**:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6953f270-6a59-4c73-aad9-0284580adccd",
"metadata": {},
"outputs": [],
"source": [
"openai_api = OpenAI()\n",
"claude_api = anthropic.Anthropic()\n",
"# ollama could be used like this:\n",
"# ollama_api = OpenAI(base_url=\"http://localhost:11434/v1\", api_key=\"ollama\")\n",
"\n",
"claude_model_str = \"claude-3-haiku-20240307\"\n",
"gpt_model_str = \"gpt-4o-mini\"\n",
"# llama_model_str = \"llama3.2\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2fadb8db-41e6-4362-a2fe-3e0902ff7116",
"metadata": {},
"outputs": [],
"source": [
"# Create Model objects\n",
"gpt_model = Model(openai_api, gpt_model_str)\n",
"claude_model = Model(claude_api, claude_model_str)\n",
"\n",
"# Create three Participants\n",
"p1 = Participant(name=name1, model=gpt_model, system_prompt=system1, initial_message=\"Hello there\")\n",
"p2 = Participant(name=name2, model=claude_model, system_prompt=system2, initial_message=\"Good evening.\")\n",
"p3 = Participant(name=name3, model=gpt_model, system_prompt=system3, initial_message=\"Hey guys\")\n",
"\n",
"# To make a mock run without API calls:\n",
"# p1 = Participant(name=name1, system_prompt=system1, initial_message=\"Hello there\")\n",
"# p2 = Participant(name=name2, system_prompt=system2, initial_message=\"Good evening.\")\n",
"# p3 = Participant(name=name3, system_prompt=system3, initial_message=\"Hey guys\")\n",
"\n",
"# Create Chat\n",
"chat = ThreeWayChat((p1, p2, p3))"
]
},
{
"cell_type": "markdown",
"id": "7f0daa3e-b97e-48ad-aa24-bff728234241",
"metadata": {},
"source": [
"#### Start the conversation:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4b377d50-52a1-4f3e-a7ed-bdc8a6abe710",
"metadata": {},
"outputs": [],
"source": [
"chat.start() # starts a chat with 4 rounds\n",
"# chat.start(2) # 2 rounds"
]
}
],
"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
}
|