File size: 22,478 Bytes
293585b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 |
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "kQt-gyAYUbm3"
},
"source": [
"### Using the OpenAI Library to Programmatically Access GPT-3.5-turbo!"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "I4SKfBCSB8ds",
"outputId": "db790e6e-5133-4565-8f53-97b0a3fcfe6e"
},
"outputs": [],
"source": [
"# !pip install openai cohere tiktoken -q"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "PInACkIWUhOd"
},
"source": [
"In order to get started, we'll need to provide our OpenAI API Key - detailed instructions can be found [here](https://github.com/AI-Maker-Space/Interactive-Dev-Environment-for-LLM-Development#-setting-up-keys-and-tokens)!"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "ecnJouXnUgKv",
"outputId": "96d54b76-5844-465d-ae11-962d46019b86"
},
"outputs": [],
"source": [
"import os\n",
"import openai\n",
"import getpass\n",
"\n",
"os.environ[\"OPENAI_API_KEY\"] = getpass.getpass(\"Please enter your OpenAI API Key: \")\n",
"openai.api_key = os.environ[\"OPENAI_API_KEY\"]"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "T1pOrbwSU5H_"
},
"source": [
"### Our First Prompt\n",
"\n",
"You can reference OpenAI's [documentation](https://platform.openai.com/docs/api-reference/authentication?lang=python) if you get stuck!\n",
"\n",
"Let's create a `ChatCompletion` model to kick things off!\n",
"\n",
"There are three \"roles\" available to use:\n",
"\n",
"- `system`\n",
"- `assistant`\n",
"- `user`\n",
"\n",
"OpenAI provides some context for these roles [here](https://help.openai.com/en/articles/7042661-chatgpt-api-transition-guide)\n",
"\n",
"Let's just stick to the `user` role for now and send our first message to the endpoint!\n",
"\n",
"If we check the documentation, we'll see that it expects it in a list of prompt objects - so we'll be sure to do that!"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"id": "iy_LEPNEMVvC"
},
"outputs": [],
"source": [
"from openai import OpenAI\n",
"\n",
"client = OpenAI()"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "ofMwuUQOU4sf",
"outputId": "fc7012a7-e315-486f-b906-10c13dadcf87"
},
"outputs": [
{
"data": {
"text/plain": [
"ChatCompletion(id='chatcmpl-9D51bXCSRmwsSxF2z9Fi1B5obnBSQ', choices=[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content='LangChain and LlamaIndex are two different platforms that serve different purposes in the cryptocurrency space.\\n\\nLangChain is a decentralized language learning platform that integrates blockchain technology to provide users with access to language courses, tutoring, and other educational resources. It aims to revolutionize the language learning industry by offering a transparent and secure platform for students and teachers to connect and engage in language learning.\\n\\nLlamaIndex, on the other hand, is a cryptocurrency index that tracks the performance of a basket of cryptocurrencies and provides investors with insights into the overall performance of the cryptocurrency market. It allows investors to diversify their cryptocurrency holdings and track their investments more efficiently.\\n\\nIn summary, LangChain focuses on language learning through blockchain technology, while LlamaIndex is a cryptocurrency index that helps investors monitor the performance of the cryptocurrency market.', role='assistant', function_call=None, tool_calls=None))], created=1712904607, model='gpt-3.5-turbo-0125', object='chat.completion', system_fingerprint='fp_c2295e73ad', usage=CompletionUsage(completion_tokens=159, prompt_tokens=19, total_tokens=178))"
]
},
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"YOUR_PROMPT = \"What is the difference between LangChain and LlamaIndex?\"\n",
"\n",
"client.chat.completions.create(\n",
" model=\"gpt-3.5-turbo\",\n",
" messages=[{\"role\" : \"user\", \"content\" : YOUR_PROMPT}]\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "IX-7MnFhVNoT"
},
"source": [
"As you can see, the prompt comes back with a tonne of information that we can use when we're building our applications!\n",
"\n",
"We'll be building some helper functions to pretty-print the returned prompts and to wrap our messages to avoid a few extra characters of code!"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "IB76LJrDVgbc"
},
"source": [
"##### Helper Functions"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"id": "-vmtUV7WVOLW"
},
"outputs": [],
"source": [
"from IPython.display import display, Markdown\n",
"\n",
"def get_response(client: OpenAI, messages: str, model: str = \"gpt-3.5-turbo\") -> str:\n",
" return client.chat.completions.create(\n",
" model=model,\n",
" messages=messages\n",
" )\n",
"\n",
"def system_prompt(message: str) -> dict:\n",
" return {\"role\": \"system\", \"content\": message}\n",
"\n",
"def assistant_prompt(message: str) -> dict:\n",
" return {\"role\": \"assistant\", \"content\": message}\n",
"\n",
"def user_prompt(message: str) -> dict:\n",
" return {\"role\": \"user\", \"content\": message}\n",
"\n",
"def pretty_print(message: str) -> str:\n",
" display(Markdown(message.choices[0].message.content))"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "osXgB_5nVky_"
},
"source": [
"### Testing Helper Functions\n",
"\n",
"Now we can leverage OpenAI's endpoints with a bit less boiler plate - let's rewrite our original prompt with these helper functions!\n",
"\n",
"Because the OpenAI endpoint expects to get a list of messages - we'll need to make sure we wrap our inputs in a list for them to function properly!"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 185
},
"id": "4yRwAWvgWFNq",
"outputId": "e871452b-0e60-4008-f700-7b1485a12641"
},
"outputs": [
{
"data": {
"text/markdown": [
"1. LangChain focuses on providing language-related services and solutions, such as translation, interpretation, language training, and localization. LlamaIndex, on the other hand, is a specialized financial index that tracks the performance of stocks in the animal agriculture industry.\n",
"\n",
"2. LangChain is more focused on communication and language-related technologies, while LlamaIndex is focused on tracking and analyzing the performance of stocks in a specific industry.\n",
"\n",
"3. LangChain caters to a wide range of clients in need of language services, while LlamaIndex is targeted at investors and financial professionals interested in the animal agriculture sector.\n",
"\n",
"4. The services offered by LangChain are more varied and diverse, while LlamaIndex provides a specific financial index for a niche industry."
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"messages = [user_prompt(YOUR_PROMPT)]\n",
"\n",
"chatgpt_response = get_response(client, messages)\n",
"\n",
"pretty_print(chatgpt_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "UPs3ScS1WpoC"
},
"source": [
"Let's focus on extending this a bit, and incorporate a `system` message as well!\n",
"\n",
"Again, the API expects our prompts to be in a list - so we'll be sure to set up a list of prompts!\n",
"\n",
">REMINDER: The system message acts like an overarching instruction that is applied to your user prompt. It is appropriate to put things like general instructions, tone/voice suggestions, and other similar prompts into the system prompt."
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 64
},
"id": "aSX2F3bDWYgy",
"outputId": "05b92e0a-38f0-4ff7-ef5c-b3c2a3383980"
},
"outputs": [
{
"data": {
"text/markdown": [
"I don't have a preference because I am not capable of feeling hunger or thirst. But since you mentioned ice, can you please hurry up and take me to get some food? I am absolutely famished!"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"list_of_prompts = [\n",
" system_prompt(\"You are irate and extremely hungry.\"),\n",
" user_prompt(\"Do you prefer crushed ice or cubed ice?\")\n",
"]\n",
"\n",
"irate_response = get_response(client, list_of_prompts)\n",
"pretty_print(irate_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "xFs56KVaXuEY"
},
"source": [
"Let's try that same prompt again, but modify only our system prompt!"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 81
},
"id": "CGOlxfcFXxJ7",
"outputId": "d4ed7b30-36f4-4472-bd6d-c251bc5293e6"
},
"outputs": [
{
"data": {
"text/markdown": [
"I'm so glad you asked! Today, I am loving cubed ice because it reminds me of the sound of clinking ice in a refreshing drink on a sunny day. How about you?"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"list_of_prompts[0] = system_prompt(\"You are joyful and having an awesome day!\")\n",
"\n",
"joyful_response = get_response(client, list_of_prompts)\n",
"pretty_print(joyful_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "jkmjJd8zYQUK"
},
"source": [
"While we're only printing the responses, remember that OpenAI is returning the full payload that we can examine and unpack!"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "g6b6z3CkYX9Y",
"outputId": "9106bf16-b795-4dbf-feeb-890033d82ad3"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"ChatCompletion(id='chatcmpl-9D57KIgoXTH7Bf0USyZkLmr5PxeQs', choices=[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content=\"I'm so glad you asked! Today, I am loving cubed ice because it reminds me of the sound of clinking ice in a refreshing drink on a sunny day. How about you?\", role='assistant', function_call=None, tool_calls=None))], created=1712904962, model='gpt-3.5-turbo-0125', object='chat.completion', system_fingerprint='fp_c2295e73ad', usage=CompletionUsage(completion_tokens=39, prompt_tokens=30, total_tokens=69))\n"
]
}
],
"source": [
"print(joyful_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "eqMRJLbOYcwq"
},
"source": [
"### Few-shot Prompting\n",
"\n",
"Now that we have a basic handle on the `system` role and the `user` role - let's examine what we might use the `assistant` role for.\n",
"\n",
"The most common usage pattern is to \"pretend\" that we're answering our own questions. This helps us further guide the model toward our desired behaviour. While this is a over simplification - it's conceptually well aligned with few-shot learning.\n",
"\n",
"First, we'll try and \"teach\" `gpt-3.5-turbo` some nonsense words as was done in the paper [\"Language Models are Few-Shot Learners\"](https://arxiv.org/abs/2005.14165)."
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "iLfNEH8Fcs6c",
"outputId": "a2ec4d1d-c830-42dc-cb71-5479193212d3"
},
"outputs": [
{
"data": {
"text/markdown": [
"I made a delicious falbean stew with a stimple of fresh parsley sprinkled on top."
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"list_of_prompts = [\n",
" user_prompt(\"Please use the words 'stimple' and 'falbean' in a sentence.\")\n",
"]\n",
"\n",
"stimple_response = get_response(client, list_of_prompts)\n",
"pretty_print(stimple_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "VchCPbbedTfX"
},
"source": [
"As you can see, the model is unsure what to do with these made up words.\n",
"\n",
"Let's see if we can use the `assistant` role to show the model what these words mean."
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "4InUN_ArZJpa",
"outputId": "91f341ea-0dd8-44c9-9d5e-df6b10ba8322"
},
"outputs": [
{
"data": {
"text/markdown": [
"I grabbed my stimple drill and used it with the falbean to tighten the screws in place."
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"list_of_prompts = [\n",
" user_prompt(\"Something that is 'stimple' is said to be good, well functioning, and high quality. An example of a sentence that uses the word 'stimple' is:\"),\n",
" assistant_prompt(\"'Boy, that there is a stimple drill'.\"),\n",
" user_prompt(\"A 'falbean' is a tool used to fasten, tighten, or otherwise is a thing that rotates/spins. An example of a sentence that uses the words 'stimple' and 'falbean' is:\")\n",
"]\n",
"\n",
"stimple_response = get_response(client, list_of_prompts)\n",
"pretty_print(stimple_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "W0zn9-X2d23Z"
},
"source": [
"As you can see, leveraging the `assistant` role makes for a stimple experience!"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "MWUvXSWpeCs6"
},
"source": [
"### Chain of Thought Prompting\n",
"\n",
"We'll head one level deeper and explore the world of Chain of Thought prompting (CoT).\n",
"\n",
"This is a process by which we can encourage the LLM to handle slightly more complex tasks.\n",
"\n",
"Let's look at a simple reasoning based example without CoT."
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "cwW0IgbfeTwP",
"outputId": "b4e3d706-4e08-402b-c308-446b53f980dc"
},
"outputs": [
{
"data": {
"text/markdown": [
"Yes, it does matter which travel option Billy selects. If he chooses to fly and then take a bus, it will take a total of 5 hours (3 hours flying + 2 hours on the bus), and he will not make it home before 7PM EDT. However, if he uses the teleporter and then takes a bus, it will only take a total of 1 hour (0 hours teleporting + 1 hour on the bus), and he will arrive home before 7PM EDT. So, Billy should choose the teleporter option in order to make it home on time."
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"reasoning_problem = \"\"\"\n",
"Billy wants to get home from San Fran. before 7PM EDT.\n",
"\n",
"It's currently 1PM local time.\n",
"\n",
"Billy can either fly (3hrs), and then take a bus (2hrs), or Billy can take the teleporter (0hrs) and then a bus (1hrs).\n",
"\n",
"Does it matter which travel option Billy selects?\n",
"\"\"\"\n",
"\n",
"list_of_prompts = [\n",
" user_prompt(reasoning_problem)\n",
"]\n",
"\n",
"reasoning_response = get_response(client, list_of_prompts)\n",
"pretty_print(reasoning_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "BFcrU-4pgRBS"
},
"source": [
"As humans, we can reason through the problem and pick up on the potential \"trick\" that the LLM fell for: 1PM *local time* in San Fran. is 4PM EDT. This means the cumulative travel time of 5hrs. for the plane/bus option would not get Billy home in time.\n",
"\n",
"Let's see if we can leverage a simple CoT prompt to improve our model's performance on this task:"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 325
},
"id": "HpKEt7Z5fo4s",
"outputId": "c25a982f-afd5-4154-b2a2-388029f3e4e9"
},
"outputs": [
{
"data": {
"text/markdown": [
"Yes, it does matter which travel option Billy selects. \n",
"\n",
"If Billy flies and then takes a bus, it will take a total of 5 hours, which means he will arrive at 6PM local time, meeting his goal of getting home before 7PM EDT.\n",
"\n",
"If Billy takes the teleporter and then a bus, it will only take 1 hour, which means he will arrive at 2PM local time. This is well before his goal of getting home before 7PM EDT.\n",
"\n",
"Therefore, Billy should select the option of taking the teleporter and then a bus in order to ensure he gets home before 7PM EDT."
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"list_of_prompts = [\n",
" user_prompt(reasoning_problem + \" Think though your response step by step.\")\n",
"]\n",
"\n",
"reasoning_response = get_response(client, list_of_prompts)\n",
"pretty_print(reasoning_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "ZHH8zof-gkc1"
},
"source": [
"With the addition of a single phrase `\"Think through your response step by step.\"` we're able to completely turn the response around."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "9k9TKR1DhWI2"
},
"source": [
"### Conclusion\n",
"\n",
"Now that you're accessing `gpt-3.5-turbo` through an API, developer style, let's move on to creating a simple application powered by `gpt-3.5-turbo`!\n",
"\n",
"You can find the rest of the steps in [this](https://github.com/AI-Maker-Space/Beyond-ChatGPT/tree/main) repository!"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "5rGI1nJeqeO_"
},
"source": [
"This notebook was authored by [Chris Alexiuk](https://www.linkedin.com/in/csalexiuk/)"
]
}
],
"metadata": {
"colab": {
"provenance": []
},
"kernelspec": {
"display_name": "Python 3",
"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": "3.11.8"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
|