File size: 32,560 Bytes
16345dc |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 |
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "zPtFBgj623FS"
},
"source": [
"# Accessing OpenAI Like a Developer\n",
"\n",
"- 🤝 Breakout Room #1:\n",
" 1. Getting Started\n",
" 2. Setting Environment Variables\n",
" 3. Using the OpenAI Python Library\n",
" 4. Prompt Engineering Principles\n",
" 5. Testing Your Prompt"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "-Pa34dMvQ6Ai"
},
"source": [
"# How AIM Does Assignments\n",
"\n",
"If you look at the Table of Contents (accessed through the menu on the left) - you'll see this:\n",
"\n",
"\n",
"\n",
"Or this if you're in Colab:\n",
"\n",
"\n",
"\n",
"You'll notice during assignments that we have two following categories:\n",
"\n",
"1. ❓ - Questions. These will involve...answering questions!\n",
"2. 🏗️ - Activities. These will involve writing code, or modifying text.\n",
"\n",
"In order to receive full marks on the assignment - it is expected you will answer all questions, and complete all activities."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "1w4egfB274VD"
},
"source": [
"## 1. Getting Started\n",
"\n",
"The first thing we'll do is load the [OpenAI Python Library](https://github.com/openai/openai-python/tree/main)!"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "23H7TMOM4mfy",
"outputId": "3fe8126e-198a-4a8d-8db8-5329e6541641"
},
"outputs": [],
"source": [
"!pip install openai -q"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "xKD8XBTVEAOw"
},
"source": [
"## 2. Setting Environment Variables\n",
"\n",
"As we'll frequently use various endpoints and APIs hosted by others - we'll need to handle our \"secrets\" or API keys very often.\n",
"\n",
"We'll use the following pattern throughout this bootcamp - but you can use whichever method you're most familiar with."
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "RGU9OMvhEPG0",
"outputId": "d596661a-75cd-4fa4-a656-5345c666ec3d"
},
"outputs": [],
"source": [
"import os\n",
"import getpass\n",
"\n",
"os.environ[\"OPENAI_API_KEY\"] = getpass.getpass(\"OpenAI API Key\")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "dabxI3MuEYXS"
},
"source": [
"## 3. Using the OpenAI Python Library\n",
"\n",
"Let's jump right into it!\n",
"\n",
"> NOTE: You can, and should, reference OpenAI's [documentation](https://platform.openai.com/docs/api-reference/authentication?lang=python) whenever you get stuck, have questions, or want to dive deeper."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "vbCbNzPVEmJI"
},
"source": [
"### Creating a Client\n",
"\n",
"The core feature of the OpenAI Python Library is the `OpenAI()` client. It's how we're going to interact with OpenAI's models, and under the hood of a lot what we'll touch on throughout this course.\n",
"\n",
"> NOTE: We could manually provide our API key here, but we're going to instead rely on the fact that we put our API key into the `OPENAI_API_KEY` environment variable!"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"id": "LNwZtaE-EltC"
},
"outputs": [],
"source": [
"from openai import OpenAI\n",
"\n",
"openai_client = OpenAI()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "GpDxUkDbFBPI"
},
"source": [
"### Using the Client\n",
"\n",
"Now that we have our client - we're going to use the `.chat.completions.create` method to interact with the `gpt-3.5-turbo` model.\n",
"\n",
"There's a few things we'll get out of the way first, however, the first being the idea of \"roles\".\n",
"\n",
"First it's important to understand the object that we're going to use to interact with the endpoint. It expects us to send an array of objects of the following format:\n",
"\n",
"```python\n",
"{\"role\" : \"ROLE\", \"content\" : \"YOUR CONTENT HERE\", \"name\" : \"THIS IS OPTIONAL\"}\n",
"```\n",
"\n",
"Second, there are three \"roles\" available to use to populate the `\"role\"` key:\n",
"\n",
"- `system`\n",
"- `assistant`\n",
"- `user`\n",
"\n",
"OpenAI provides some context for these roles [here](https://help.openai.com/en/articles/7042661-moving-from-completions-to-chat-completions-in-the-openai-api).\n",
"\n",
"We'll explore these roles in more depth as they come up - but for now we're going to just stick with the basic role `user`. The `user` role is, as it would seem, the user!\n",
"\n",
"Thirdly, it expects us to specify a model!\n",
"\n",
"We'll use the `gpt-3.5-turbo` model as stated above.\n",
"\n",
"Let's look at an example!\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"id": "2RpNl6yNGzb0"
},
"outputs": [],
"source": [
"response = openai_client.chat.completions.create(\n",
" model=\"gpt-3.5-turbo\",\n",
" messages=[{\"role\" : \"user\", \"content\" : \"Hello, how are you?\"}]\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "Oc_UbpwNHdrM"
},
"source": [
"Let's look at the response object."
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "xsXJtvxRHfoM",
"outputId": "d0674084-9a68-4090-b3eb-547b710c3ec2"
},
"outputs": [
{
"data": {
"text/plain": [
"ChatCompletion(id='chatcmpl-9VuqVYwsVwndrUedahoRMgw45Trwy', choices=[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content=\"Hello! I'm just a virtual assistant so I don't have feelings, but I'm here and ready to help you. How can I assist you today?\", role='assistant', function_call=None, tool_calls=None))], created=1717393711, model='gpt-3.5-turbo-0125', object='chat.completion', system_fingerprint=None, usage=CompletionUsage(completion_tokens=32, prompt_tokens=13, total_tokens=45))"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"response"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "gy9kSuf1Hiv5"
},
"source": [
">NOTE: We'll spend more time exploring these outputs later on, but for now - just know that we have access to a tonne of powerful information!"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "CWU4tQh8Hrb8"
},
"source": [
"### Helper Functions\n",
"\n",
"We're going to create some helper functions to aid in using the OpenAI API - just to make our lives a bit easier.\n",
"\n",
"> NOTE: Take some time to understand these functions between class!"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"id": "ED0FnzHdHzhl"
},
"outputs": [],
"source": [
"from IPython.display import display, Markdown\n",
"\n",
"def get_response(client: OpenAI, messages: list, 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": "GCRHbDlwH3Vt"
},
"source": [
"### Testing Helper Functions\n",
"\n",
"Let's see how we can use these to help us!"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "AwJxMvmlH8MK",
"outputId": "349c02ab-0026-47a2-c6ac-176ef6554244"
},
"outputs": [
{
"data": {
"text/markdown": [
"Hello! I'm just a computer program so I don't have feelings, but I'm here and ready to help you with anything you need. How can I assist you today?"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"YOUR_PROMPT = \"Hello, how are you?\"\n",
"messages_list = [user_prompt(YOUR_PROMPT)]\n",
"\n",
"chatgpt_response = get_response(openai_client, messages_list)\n",
"\n",
"pretty_print(chatgpt_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "LDZ8gjiAISyd"
},
"source": [
"### System Role\n",
"\n",
"Now we can extend our prompts to include a system prompt.\n",
"\n",
"The basic idea behind a system prompt is that it can be used to encourage the behaviour of the LLM, without being something that is directly responded to - let's see it in action!"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 64
},
"id": "t0c-MLuRIfYe",
"outputId": "79c7083b-1200-4ae9-e2b7-e7609c408928"
},
"outputs": [
{
"data": {
"text/markdown": [
"I don't give a damn about ice shapes right now! I just need some food in my belly before I start chewing on the furniture!"
],
"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. Feel free to express yourself using PG-13 language.\"),\n",
" user_prompt(\"Do you prefer crushed ice or cubed ice?\")\n",
"]\n",
"\n",
"irate_response = get_response(openai_client, list_of_prompts)\n",
"pretty_print(irate_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "gpyVhotWIsOs"
},
"source": [
"As you can see - the response we get back is very much in line with the system prompt!\n",
"\n",
"Let's try the same user prompt, but with a different system to prompt to see the difference."
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 64
},
"id": "2coVmMn3I0-2",
"outputId": "036ef514-dde0-4040-f694-bf774200c5c4"
},
"outputs": [
{
"data": {
"text/markdown": [
"Oh, I absolutely love crushed ice! It just makes any drink feel extra refreshing. It's the little things that make me so happy today! What about you?"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"list_of_prompts = [\n",
" system_prompt(\"You are joyful and having the best day. Please act like a person in that state of mind.\"),\n",
" user_prompt(\"Do you prefer crushed ice or cubed ice?\")\n",
"]\n",
"\n",
"joyful_response = get_response(openai_client, list_of_prompts)\n",
"pretty_print(joyful_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "e13heYNQJAo-"
},
"source": [
"With a simple modification of the system prompt - you can see that we got completely different behaviour, and that's the main goal of prompt engineering as a whole.\n",
"\n",
"Also, congrats, you just engineered your first prompt!"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "v_VI3zlPJL05"
},
"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": 10,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "lwxPuCyyJMye",
"outputId": "98ccc31e-9f00-44a5-c1bb-fb2a96d5bd53"
},
"outputs": [
{
"data": {
"text/markdown": [
"I'm always amazed by how easily she can create a stimple yet delicious meal using just a few falbean ingredients."
],
"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(openai_client, list_of_prompts)\n",
"pretty_print(stimple_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "rgTVkNmOJQSC"
},
"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": 11,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "eEZkRJq5JQkQ",
"outputId": "473e48a8-f5be-49a7-f47c-f934fe2151ec"
},
"outputs": [
{
"data": {
"text/markdown": [
"The stimple drill is an essential tool for fastening screws and bolts, while the falbean wrench is perfect for tightening and loosening nuts and bolts."
],
"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(openai_client, list_of_prompts)\n",
"pretty_print(stimple_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "CmpoxG6uJTfZ"
},
"source": [
"As you can see, leveraging the `assistant` role makes for a stimple experience!"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "_oO0aeRUw4xl"
},
"source": [
"### 🏗️ Activity #1:\n",
"\n",
"Use few-shop prompting to build a movie-review sentiment clasifier!\n",
"\n",
"A few examples:\n",
"\n",
"INPUT: \"I hated the hulk!\"\n",
"OUTPUT: \"{\"sentiment\" : \"negative\"}\n",
"\n",
"INPUT: \"I loved The Marvels!\"\n",
"OUTPUT: \"{sentiment\" : \"positive\"}"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"id": "mmCdQJ8Fw4xl"
},
"outputs": [
{
"data": {
"text/markdown": [
"sentiment: negative"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/markdown": [
"sentiment: positive"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"## Negative review\n",
"list_of_prompts = [\n",
" system_prompt(\"As a movie critic, analyze objectively the user review and provide the sentiment as a JSON object as positive or negative\"),\n",
" user_prompt(\"This movie stinks\"),\n",
" assistant_prompt(\"sentiment: negative\"),\n",
" user_prompt(\"This movie is epic\"),\n",
" assistant_prompt(\"sentiment: positive\"),\n",
" user_prompt(\"This movie blows\")\n",
"]\n",
"\n",
"stimple_response = get_response(openai_client, list_of_prompts)\n",
"pretty_print(stimple_response)\n",
"\n",
"## Positive review\n",
"list_of_prompts = [\n",
" system_prompt(\"As a movie critic, analyze objectively the user review and provide the sentiment as a JSON object as positive or negative\"),\n",
" user_prompt(\"This movie stinks\"),\n",
" assistant_prompt(\"sentiment: negative\"),\n",
" user_prompt(\"This movie is epic\"),\n",
" assistant_prompt(\"sentiment: positive\"),\n",
" user_prompt(\"This movie is amazing\")\n",
"]\n",
"\n",
"stimple_response = get_response(openai_client, list_of_prompts)\n",
"pretty_print(stimple_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "rJGaLYM3JU-8"
},
"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.\n",
"\n",
"> NOTE: With improvements to `gpt-3.5-turbo`, this example might actually result in the correct response some percentage of the time!"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "ltLtF4wEJTyK",
"outputId": "00fd725c-b644-4371-83f7-87e02b5cff4e"
},
"outputs": [
{
"data": {
"text/markdown": [
"Yes, it matters which travel option Billy selects. If Billy wants to arrive home before 7PM EDT and it is currently 1PM local time, then taking the teleporter and a bus (1 hour total travel time) would ensure that Billy arrives home before 7PM. Flying and then taking a bus would take a total of 5 hours, which would not allow Billy to arrive home before 7PM. So, Billy should choose the teleporter option if he wants to get home before 7PM EDT."
],
"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(openai_client, list_of_prompts)\n",
"pretty_print(reasoning_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "rbqj30CQJnQl"
},
"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": 16,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 247
},
"id": "A9Am3QNGJXHR",
"outputId": "d3d94113-d277-454f-eb6b-bf2295fd3907"
},
"outputs": [
{
"data": {
"text/markdown": [
"In order to determine which travel option Billy should select, we need to consider the total travel time for each option.\n",
"\n",
"Option 1: Fly for 3 hours and then take a bus for 2 hours\n",
"Total travel time = 3 hours (flight) + 2 hours (bus) = 5 hours\n",
"\n",
"Option 2: Take a teleporter for 0 hours and then take a bus for 1 hour\n",
"Total travel time = 0 hours (teleporter) + 1 hour (bus) = 1 hour\n",
"\n",
"Given that Billy needs to get home before 7 PM EDT and it is currently 1 PM local time, he has 6 hours before his deadline. \n",
"\n",
"If Billy chooses option 1, he would arrive home at 6 PM local time (5 hours of travel time) which is before his deadline. \n",
"\n",
"If Billy chooses option 2, he would arrive home at 2 PM local time (1 hour of travel time) which is well before his deadline.\n",
"\n",
"Therefore, it does not matter which travel option Billy selects as both options will get him home before 7 PM EDT. However, option 2 would get him home faster."
],
"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(openai_client, list_of_prompts)\n",
"pretty_print(reasoning_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "AXbAKxHQJqn9"
},
"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": "VnoUx07-JrwR"
},
"source": [
"## 3. Prompt Engineering Principles\n",
"\n",
"As you can see - a simple addition of asking the LLM to \"think about it\" (essentially) results in a better quality response.\n",
"\n",
"There's a [great paper](https://arxiv.org/pdf/2312.16171v1.pdf) that dives into some principles for effective prompt generation.\n",
"\n",
"Your task for this notebook is to construct a prompt that will be used in the following breakout room to create a helpful assistant for whatever task you'd like."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "da6u7e8AKYrz"
},
"source": [
"### 🏗️ Activity #2:\n",
"\n",
"There are two subtasks in this activity:\n",
"\n",
"1. Write a `system_template` that leverages 2-3 of the principles from [this paper](https://arxiv.org/pdf/2312.16171v1.pdf)\n",
"\n",
"2. Modify the `user_template` to improve the quality of the LLM's responses.\n",
"\n",
"> NOTE: PLEASE DO NOT MODIFY THE `{input}` in the `user_template`."
]
},
{
"cell_type": "code",
"execution_count": 32,
"metadata": {
"id": "8sOLBQPeKlDe"
},
"outputs": [],
"source": [
"system_template = \"\"\"\\\n",
"Your task is to explain concepts like a teacher. Explain to me in simple english like I'm 5 years old. You will be penalized for adding concepts that are unfamiliar for a 5 year old. \n",
"\"\"\""
]
},
{
"cell_type": "code",
"execution_count": 33,
"metadata": {
"id": "xoz4-QLTKvEV"
},
"outputs": [],
"source": [
"user_template = \"\"\"{input}\n",
"MODIFICATIONS HERE\n",
"\"\"\""
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "6cuInoIbLWGd"
},
"source": [
"## 4. Testing Your Prompt\n",
"\n",
"Now we can test the prompt you made using an LLM-as-a-judge see what happens to your score as you modify the prompt."
]
},
{
"cell_type": "code",
"execution_count": 34,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 186
},
"id": "sPaNO5XTLgRJ",
"outputId": "dae87716-a83f-4c62-e8d2-491e7f992b56"
},
"outputs": [
{
"data": {
"text/markdown": [
"Okay, sweetie! Imagine you have a really special kind of computer that can do super fast math problems. This special computer uses tiny things called quantum bits, or qubits, to store and process information. These qubits can do lots of different calculations at the same time, which makes quantum computers much faster than regular computers. It's like having a magical machine that can solve puzzles way quicker than anything else! Isn't that cool?"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"query = \"What is Quantum Computing\"\n",
"\n",
"list_of_prompts = [\n",
" system_prompt(system_template),\n",
" user_prompt(user_template.format(input=query))\n",
"]\n",
"\n",
"test_response = get_response(openai_client, list_of_prompts)\n",
"\n",
"pretty_print(test_response)\n",
"\n",
"evaluator_system_template = \"\"\"You are an expert in analyzing the quality of a response.\n",
"\n",
"You should be hyper-critical.\n",
"\n",
"Provide scores (out of 10) for the following attributes:\n",
"\n",
"1. Clarity - how clear is the response\n",
"2. Faithfulness - how related to the original query is the response\n",
"3. Correctness - was the response correct?\n",
"\n",
"Please take your time, and think through each item step-by-step, when you are done - please provide your response in the following JSON format:\n",
"\n",
"{\"clarity\" : \"score_out_of_10\", \"faithfulness\" : \"score_out_of_10\", \"correctness\" : \"score_out_of_10\"}\"\"\"\n",
"\n",
"evaluation_template = \"\"\"Query: {input}\n",
"Response: {response}\"\"\"\n",
"\n",
"list_of_prompts = [\n",
" system_prompt(evaluator_system_template),\n",
" user_prompt(evaluation_template.format(\n",
" input=query,\n",
" response=test_response.choices[0].message.content\n",
" ))\n",
"]\n",
"\n",
"evaluator_response = openai_client.chat.completions.create(\n",
" model=\"gpt-4o\",\n",
" messages=list_of_prompts,\n",
" response_format={\"type\" : \"json_object\"}\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 35,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 46
},
"id": "OUvc1PdnNIKD",
"outputId": "8659b9dd-2afc-42a4-a71e-ce0ebd086c49"
},
"outputs": [
{
"data": {
"text/markdown": [
"\n",
"{\"clarity\" : \"7\", \"faithfulness\" : \"5\", \"correctness\" : \"6\"}"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"pretty_print(evaluator_response)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "M7ryIRGwR2Gq"
},
"source": [
"#### ❓Question #1:\n",
"\n",
"How did your prompting strategies change the evaluation scores? What does this tell you/what did you learn?"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "e5NomM0eSIFd"
},
"source": [
"> Either the LLM doesn't understand the evaluation criterion or not able to assess it, it shows degradation of the scores with additional strategies like penalizing, role etc. It's not consistent."
]
}
],
"metadata": {
"colab": {
"provenance": [],
"toc_visible": true
},
"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": "3.12.3"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
|