File size: 48,799 Bytes
611d08c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 |
{
"cells": [
{
"cell_type": "markdown",
"id": "873f35d6-df57-48e0-a60a-45d3c9e15c96",
"metadata": {},
"source": [
"# Bonito: Synthetic Data Generation"
]
},
{
"cell_type": "markdown",
"id": "c283bd6d-ea2f-4a6f-99ac-8887e7621bcb",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:32:57.354413Z",
"iopub.status.busy": "2024-09-17T08:32:57.354090Z",
"iopub.status.idle": "2024-09-17T08:32:57.357571Z",
"shell.execute_reply": "2024-09-17T08:32:57.356909Z",
"shell.execute_reply.started": "2024-09-17T08:32:57.354388Z"
}
},
"source": [
"#### Environment Setup"
]
},
{
"cell_type": "raw",
"id": "df65c34e-9080-4e11-9b21-12fa6f459736",
"metadata": {},
"source": [
"pip freeze >> requirements_bonito.txt"
]
},
{
"cell_type": "raw",
"id": "3502e6ea-1ce9-497a-8bfa-dfc6ff215d07",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:26:59.911435Z",
"iopub.status.busy": "2024-09-17T08:26:59.911244Z",
"iopub.status.idle": "2024-09-17T08:26:59.913754Z",
"shell.execute_reply": "2024-09-17T08:26:59.913427Z",
"shell.execute_reply.started": "2024-09-17T08:26:59.911420Z"
}
},
"source": [
"conda create -n bonito python=3.9\n",
"conda activate bonito\n",
"pip install -e ."
]
},
{
"cell_type": "raw",
"id": "0940de6a-0e05-41f0-9d72-2f206bf77ff0",
"metadata": {},
"source": [
"pip install autoawq\n",
"pip install flash-attn==2.5.6 --no-build-isolation"
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "b33c2d03-9cb9-4f31-bcae-d81941486e25",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:33:56.866768Z",
"iopub.status.busy": "2024-09-17T08:33:56.866424Z",
"iopub.status.idle": "2024-09-17T08:33:56.872656Z",
"shell.execute_reply": "2024-09-17T08:33:56.872137Z",
"shell.execute_reply.started": "2024-09-17T08:33:56.866742Z"
}
},
"outputs": [
{
"data": {
"text/plain": [
"('2.4.0+cu121', '4.44.2', '0.6.1.post2', '0.2.6')"
]
},
"execution_count": 18,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"import transformers\n",
"import torch\n",
"import vllm\n",
"import awq\n",
"\n",
"torch.__version__, transformers.__version__, vllm.__version__, awq.__version__"
]
},
{
"cell_type": "markdown",
"id": "987a88f2-2240-47ab-b0b7-d1589a6da59e",
"metadata": {},
"source": [
"#### Quantized Bonito Wrapper\n",
"This is a simplified quantized bonito class to generate a single synthetic input-output instruction for a given text and task type.\n",
"This code uses huggingface `transformers` library for generation.\n",
"For complete functionality and faster generations, we recommend using the `Bonito` class from the package."
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "20855b3e-e864-4875-9135-bf1242a99242",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:26:25.162113Z",
"iopub.status.busy": "2024-09-17T08:26:25.161926Z",
"iopub.status.idle": "2024-09-17T08:26:25.167686Z",
"shell.execute_reply": "2024-09-17T08:26:25.166893Z",
"shell.execute_reply.started": "2024-09-17T08:26:25.162099Z"
}
},
"outputs": [],
"source": [
"from typing import Dict, Optional, Union\n",
"from datasets import Dataset\n",
"\n",
"SHORTFORM_TO_FULL_TASK_TYPES = {\n",
" \"exqa\": \"extractive question answering\",\n",
" \"mcqa\": \"multiple-choice question answering\",\n",
" \"qg\": \"question generation\",\n",
" \"qa\": \"question answering without choices\",\n",
" \"ynqa\": \"yes-no question answering\",\n",
" \"coref\": \"coreference resolution\",\n",
" \"paraphrase\": \"paraphrase generation\",\n",
" \"paraphrase_id\": \"paraphrase identification\",\n",
" \"sent_comp\": \"sentence completion\",\n",
" \"sentiment\": \"sentiment\",\n",
" \"summarization\": \"summarization\",\n",
" \"text_gen\": \"text generation\",\n",
" \"topic_class\": \"topic classification\",\n",
" \"wsd\": \"word sense disambiguation\",\n",
" \"te\": \"textual entailment\",\n",
" \"nli\": \"natural language inference\",\n",
"}\n",
"\n",
"class AbstractBonito:\n",
" def _prepare_bonito_input(\n",
" self, context_dataset: Dataset, task_type: str, context_col: str, **kwargs\n",
" ) -> Dataset:\n",
" \"\"\"\n",
" Prepares the input for the Bonito model.\n",
"\n",
" This method takes a context dataset, a task type, and a context\n",
" column name, and prepares the dataset for the Bonito model.\n",
" If the task type is not recognized, it raises a ValueError.\n",
"\n",
" Args:\n",
" context_dataset (Dataset): The dataset that provides the\n",
" context for the task.\n",
" task_type (str): The type of the task. This can be a\n",
" short form or a full form. If the task type is not\n",
" recognized, a ValueError is raised.\n",
" context_col (str): The name of the column in the dataset\n",
" that provides the context for the task.\n",
" **kwargs: Additional keyword arguments.\n",
"\n",
" Returns:\n",
" Dataset: The prepared dataset for the Bonito model.\n",
" \"\"\"\n",
" # get the task type name\n",
" if task_type in SHORTFORM_TO_FULL_TASK_TYPES.values():\n",
" full_task_type = task_type\n",
" elif task_type in SHORTFORM_TO_FULL_TASK_TYPES:\n",
" full_task_type = SHORTFORM_TO_FULL_TASK_TYPES[task_type]\n",
" else:\n",
" raise ValueError(f\"Task type {task_type} not recognized\")\n",
"\n",
" def process(example):\n",
" input_text = \"<|tasktype|>\\n\" + full_task_type.strip()\n",
" input_text += (\n",
" \"\\n<|context|>\\n\" + example[context_col].strip() + \"\\n<|task|>\\n\"\n",
" )\n",
" return {\n",
" \"input\": input_text,\n",
" }\n",
"\n",
" return context_dataset.map(\n",
" process,\n",
" remove_columns=context_dataset.column_names,\n",
" num_proc=kwargs.get(\"num_proc\", 1),\n",
" )\n",
"\n",
" def _postprocess_dataset(\n",
" self, synthetic_dataset: Dataset, context_col: str, **kwargs\n",
" ) -> Dataset:\n",
" \"\"\"\n",
" Post-processes the synthetic dataset.\n",
"\n",
" This method takes a synthetic dataset and a context column\n",
" name, and post-processes the dataset. It filters out\n",
" examples where the prediction does not contain exactly two\n",
" parts separated by \"<|pipe|>\", and then maps each example to a\n",
" new format where the context is inserted into the first part of\n",
" the prediction and the second part of the prediction is used as\n",
" the output.\n",
"\n",
" Args:\n",
" synthetic_dataset (Dataset): The synthetic dataset to be\n",
" post-processed.\n",
" context_col (str): The name of the column in the dataset\n",
" that provides the context for the tasks.\n",
" **kwargs: Additional keyword arguments.\n",
"\n",
" Returns:\n",
" Dataset: The post-processed synthetic dataset.\n",
" \"\"\"\n",
" synthetic_dataset = synthetic_dataset.filter(\n",
" lambda example: len(example[\"prediction\"].split(\"<|pipe|>\")) == 2\n",
" )\n",
"\n",
" def process(example):\n",
" pair = example[\"prediction\"].split(\"<|pipe|>\")\n",
" context = example[context_col].strip()\n",
" return {\n",
" \"input\": pair[0].strip().replace(\"{{context}}\", context),\n",
" \"output\": pair[1].strip(),\n",
" }\n",
"\n",
" synthetic_dataset = synthetic_dataset.map(\n",
" process,\n",
" remove_columns=synthetic_dataset.column_names,\n",
" num_proc=kwargs.get(\"num_proc\", 1),\n",
" )\n",
"\n",
" return synthetic_dataset\n"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "ea7d9840-162e-4c1e-b25d-354b1e110bdc",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:19:16.539490Z",
"iopub.status.busy": "2024-09-17T08:19:16.539225Z",
"iopub.status.idle": "2024-09-17T08:19:16.807696Z",
"shell.execute_reply": "2024-09-17T08:19:16.807297Z",
"shell.execute_reply.started": "2024-09-17T08:19:16.539469Z"
}
},
"outputs": [],
"source": [
"from typing import Optional, List, Dict\n",
"from datasets import Dataset\n",
"from awq import AutoAWQForCausalLM\n",
"from transformers import AutoTokenizer\n",
"\n",
"class AWQBonito(AbstractBonito):\n",
" def __init__(self, model_name_or_path):\n",
" self.model = AutoAWQForCausalLM.from_quantized(model_name_or_path, fuse_layers=True, torch_dtype=torch.float16)\n",
" self.tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=True) \n",
" \n",
" def generate_task(\n",
" self,\n",
" unannotated_paragraph: str,\n",
" task_type: str,\n",
" sampling_params: dict,\n",
" context_col=\"input\"\n",
" ) -> Dict:\n",
" \"\"\"\n",
" Generates synthetic instruction tuning pair using the Quantized Bonito model.\n",
" This method takes a text unannotated text, a task type, and sampling parameters,\n",
" and generates synthetic input-output pair.\n",
"\n",
" Args:\n",
" unannotated_paragraph (str): The unannotated text or a paragraph\n",
" task_type (str): The type of the tasks. This can be a\n",
" short form or a full form.\n",
" sampling_params (dict): The parameters for\n",
" sampling.\n",
" **kwargs: Additional keyword arguments.\n",
"\n",
" Returns:\n",
" Dict: The synthetic input-output pair for the task type.\n",
" \"\"\"\n",
"\n",
" text_dataset = Dataset.from_list([{context_col: unannotated_paragraph}])\n",
"\n",
" processed_dataset = self._prepare_bonito_input(\n",
" text_dataset, task_type, context_col=context_col\n",
" )\n",
"\n",
" outputs = self._generate_text(processed_dataset[\"input\"], sampling_params)\n",
" examples = []\n",
" for i, example in enumerate(text_dataset.to_list()):\n",
" output = outputs[i]\n",
" example[\"prediction\"] = output.strip()\n",
" examples.append(example)\n",
"\n",
" synthetic_dataset = Dataset.from_list(examples)\n",
"\n",
" # filter out the examples that cannot be parsed\n",
" synthetic_dataset_dict = self._postprocess_dataset(\n",
" synthetic_dataset, context_col=context_col\n",
" ).to_list()[0]\n",
"\n",
" return synthetic_dataset_dict\n",
"\n",
" def _generate_text(\n",
" self,\n",
" dataset: Dataset,\n",
" sampling_params: dict,\n",
" ) -> List[str]:\n",
" \"\"\"\n",
" Generate text using huggingface transformers generate function.\n",
"\n",
" This method takes a dataset of prompts, encodes them,\n",
" generates text using the model, decodes the generated\n",
" text, and appends it to a list.\n",
"\n",
" Args:\n",
" dataset (Dataset): A dataset containing prompts for text generation.\n",
" sampling_params (dict): Parameters for sampling during generation.\n",
"\n",
" Returns:\n",
" List[str]: A list of generated texts corresponding to the prompts.\n",
" \"\"\"\n",
" generated_texts = []\n",
"\n",
" for prompt in dataset:\n",
" input_ids = self.tokenizer.encode(prompt, return_tensors=\"pt\")\n",
" input_ids = input_ids.cuda()\n",
"\n",
" output = self.model.generate(input_ids, do_sample=True, **sampling_params)\n",
"\n",
" generated_text = self.tokenizer.decode(\n",
" output[0][len(input_ids[0]) :], skip_special_tokens=True\n",
" )\n",
" generated_texts.append(generated_text)\n",
"\n",
" return generated_texts"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "7f773af6-b700-4368-aaed-6bda08560e16",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:19:17.123688Z",
"iopub.status.busy": "2024-09-17T08:19:17.123507Z",
"iopub.status.idle": "2024-09-17T08:19:17.127238Z",
"shell.execute_reply": "2024-09-17T08:19:17.126899Z",
"shell.execute_reply.started": "2024-09-17T08:19:17.123674Z"
}
},
"outputs": [],
"source": [
"from datasets import Dataset\n",
"from vllm import LLM, SamplingParams\n",
"\n",
"class VLLMBonito(LLM, AbstractBonito):\n",
" \n",
" def generate_tasks(\n",
" self,\n",
" text_dataset: Dataset,\n",
" context_col: str,\n",
" task_type: str,\n",
" sampling_params: SamplingParams,\n",
" **kwargs,\n",
" ):\n",
" \"\"\"\n",
" Generates tasks using the Bonito model.\n",
"\n",
" This method takes a text dataset, a context column name,\n",
" a task type, and sampling parameters, and generates tasks\n",
" using the Bonito model. It processes the input dataset,\n",
" generates outputs, collects multiple generations into\n",
" one dataset object, and filters out the examples that\n",
" cannot be parsed.\n",
"\n",
" Args:\n",
" text_dataset (Dataset): The dataset that provides the text\n",
" for the tasks.\n",
" context_col (str): The name of the column in the dataset\n",
" that provides the context for the tasks.\n",
" task_type (str): The type of the tasks. This can be a\n",
" short form or a full form.\n",
" sampling_params (SamplingParams): The parameters for\n",
" sampling.\n",
" **kwargs: Additional keyword arguments.\n",
"\n",
" Returns:\n",
" Dataset: The synthetic dataset with the generated tasks.\n",
" \"\"\"\n",
" processed_dataset = self._prepare_bonito_input(\n",
" text_dataset, task_type, context_col, **kwargs\n",
" )\n",
" outputs = self.generate(processed_dataset[\"input\"], sampling_params)\n",
"\n",
" # collect multiple generations into one dataset object\n",
" examples = []\n",
" for i, example in enumerate(text_dataset.to_list()):\n",
" for output in outputs[i].outputs:\n",
" examples.append(\n",
" {\"context\": example[context_col], \"prediction\": output.text.strip()}\n",
" )\n",
"\n",
" synthetic_dataset = Dataset.from_list(examples)\n",
"\n",
" # filter out the examples that cannot be parsed\n",
" synthetic_dataset = self._postprocess_dataset(\n",
" synthetic_dataset, context_col=\"context\", **kwargs\n",
" )\n",
"\n",
" return synthetic_dataset"
]
},
{
"cell_type": "markdown",
"id": "5d666ab6-7722-4afd-9bde-dad5039f442e",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T07:41:42.559750Z",
"iopub.status.busy": "2024-09-17T07:41:42.559531Z",
"iopub.status.idle": "2024-09-17T07:41:42.563349Z",
"shell.execute_reply": "2024-09-17T07:41:42.562597Z",
"shell.execute_reply.started": "2024-09-17T07:41:42.559738Z"
}
},
"source": [
"## Synthetic Data Generation\n",
"Here we will load the quantized bonito model and generate synthetic instruction for the unannotated text.\n",
"\n",
"```\n",
"SHORTFORM_TO_FULL_TASK_TYPES = {\n",
" \"exqa\": \"extractive question answering\",\n",
" \"mcqa\": \"multiple-choice question answering\",\n",
" \"qg\": \"question generation\",\n",
" \"qa\": \"question answering without choices\",\n",
" \"ynqa\": \"yes-no question answering\",\n",
" \"coref\": \"coreference resolution\",\n",
" \"paraphrase\": \"paraphrase generation\",\n",
" \"paraphrase_id\": \"paraphrase identification\",\n",
" \"sent_comp\": \"sentence completion\",\n",
" \"sentiment\": \"sentiment\",\n",
" \"summarization\": \"summarization\",\n",
" \"text_gen\": \"text generation\",\n",
" \"topic_class\": \"topic classification\",\n",
" \"wsd\": \"word sense disambiguation\",\n",
" \"te\": \"textual entailment\",\n",
" \"nli\": \"natural language inference\",\n",
"}\n",
"```"
]
},
{
"cell_type": "markdown",
"id": "a577b15a-fd75-48bb-8faf-bb97a424db48",
"metadata": {},
"source": [
"### Generate the synthetic instructions\n",
"After loading the model, we pass the unannotated paragraph and the task type to generate the instructions.\n",
"Here we generate an NLI task:"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "3704550e-e109-404c-8d0e-127bcebf28e6",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:22:59.468174Z",
"iopub.status.busy": "2024-09-17T08:22:59.467863Z",
"iopub.status.idle": "2024-09-17T08:22:59.472978Z",
"shell.execute_reply": "2024-09-17T08:22:59.472332Z",
"shell.execute_reply.started": "2024-09-17T08:22:59.468149Z"
}
},
"outputs": [
{
"data": {
"text/plain": [
"'\\n1. “Confidential Information”, whenever used in this Agreement, shall mean any data, document, specification and other information or material, that is delivered or disclosed by UNHCR to the Recipient in any form whatsoever, whether orally, visually in writing or otherwise (including computerized form), and that, at the time of disclosure to the Recipient, is designated as confidential.\\n'"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"## sample text\n",
"unannotated_paragraph = \"\"\"\n",
"1. “Confidential Information”, whenever used in this Agreement, shall mean any data, document, specification and other information or material, that is delivered or disclosed by UNHCR to the Recipient in any form whatsoever, whether orally, visually in writing or otherwise (including computerized form), and that, at the time of disclosure to the Recipient, is designated as confidential.\n",
"\"\"\"\n",
"unannotated_paragraph"
]
},
{
"cell_type": "markdown",
"id": "aa1e941d-317d-415d-b2fb-5177394f68f9",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:31:33.945680Z",
"iopub.status.busy": "2024-09-17T08:31:33.945338Z",
"iopub.status.idle": "2024-09-17T08:31:33.949066Z",
"shell.execute_reply": "2024-09-17T08:31:33.948384Z",
"shell.execute_reply.started": "2024-09-17T08:31:33.945653Z"
}
},
"source": [
"### Quantized Bonito : AWQ Inference"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "012d7f03-3629-45a0-be38-64b90dc6ce0a",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"\n",
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
"model_id = 'mychen76/Llama-3.1-8B-bonito-v1-awq'\n",
"bonito = AWQBonito(model_id)"
]
},
{
"cell_type": "markdown",
"id": "4c70d39e-f790-47c3-9b2c-ed5212cc6416",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:17:55.265931Z",
"iopub.status.busy": "2024-09-17T08:17:55.265618Z",
"iopub.status.idle": "2024-09-17T08:17:55.269425Z",
"shell.execute_reply": "2024-09-17T08:17:55.268769Z",
"shell.execute_reply.started": "2024-09-17T08:17:55.265905Z"
}
},
"source": [
"### generate nli"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ede84178-63b1-4df0-a1c5-1dfefa0a986e",
"metadata": {},
"outputs": [],
"source": [
"from transformers import set_seed\n",
"from pprint import pprint\n",
"set_seed(2)\n",
"\n",
"# Generate synthetic instruction tuning dataset\n",
"sampling_params = {\n",
" \"max_new_tokens\": 256,\n",
" \"top_p\": 0.95,\n",
" \"temperature\": 0.7,\n",
" \"num_return_sequences\": 1,\n",
"}\n",
"synthetic_dataset = bonito.generate_task(\n",
" unannotated_paragraph, task_type=\"nli\", sampling_params=sampling_params\n",
")\n",
"pprint(\"----Generated Instructions----\")\n",
"pprint(f'Input: {synthetic_dataset[\"input\"]}')\n",
"pprint(f'Output: {synthetic_dataset[\"output\"]}')"
]
},
{
"cell_type": "markdown",
"id": "a7fcf6e8-f6cb-4ac8-b37b-f193ffd6f924",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:18:08.288818Z",
"iopub.status.busy": "2024-09-17T08:18:08.288500Z",
"iopub.status.idle": "2024-09-17T08:18:08.292090Z",
"shell.execute_reply": "2024-09-17T08:18:08.291481Z",
"shell.execute_reply.started": "2024-09-17T08:18:08.288792Z"
}
},
"source": [
"### generate qa"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ef1d3ba1-5a2f-4a7b-adce-5b030df2fb73",
"metadata": {},
"outputs": [],
"source": [
"from transformers import set_seed\n",
"from pprint import pprint\n",
"set_seed(2)\n",
"\n",
"# Generate synthetic instruction tuning dataset\n",
"sampling_params = {\n",
" \"max_new_tokens\": 256,\n",
" \"top_p\": 0.95,\n",
" \"temperature\": 0.7,\n",
" \"num_return_sequences\": 1,\n",
"}\n",
"synthetic_dataset = bonito.generate_task(\n",
" unannotated_paragraph, task_type=\"qa\", sampling_params=sampling_params\n",
")\n",
"pprint(\"----Generated Instructions----\")\n",
"pprint(f'Input: {synthetic_dataset[\"input\"]}')\n",
"pprint(f'Output: {synthetic_dataset[\"output\"]}')"
]
},
{
"cell_type": "markdown",
"id": "e9f79b37-d1bc-49bf-8d7d-fc2d088846c0",
"metadata": {},
"source": [
"### vLLM Bonito : AWQ Inference\n",
"ValueError: The model's max seq len (131072) is larger than the maximum number of tokens that can be stored in KV cache (129248). Try increasing `gpu_memory_utilization` or decreasing `max_model_len` when initializing the engine."
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "1a3ffd80-94f0-47bc-b1fe-47444bda66b6",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:19:02.242069Z",
"iopub.status.busy": "2024-09-17T08:19:02.241824Z",
"iopub.status.idle": "2024-09-17T08:19:04.107622Z",
"shell.execute_reply": "2024-09-17T08:19:04.107273Z",
"shell.execute_reply.started": "2024-09-17T08:19:02.242049Z"
}
},
"outputs": [],
"source": [
"from datasets import load_dataset\n",
"\n",
"# load dataaset with unannotated text\n",
"unannotated_text_ds = load_dataset(\n",
" \"BatsResearch/bonito-experiment\",\n",
" \"unannotated_contract_nli\"\n",
")[\"train\"].select(range(10))"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "98f9a917-070d-4c10-9d29-5ecce6c786bb",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:19:22.658366Z",
"iopub.status.busy": "2024-09-17T08:19:22.658098Z",
"iopub.status.idle": "2024-09-17T08:19:39.785070Z",
"shell.execute_reply": "2024-09-17T08:19:39.784677Z",
"shell.execute_reply.started": "2024-09-17T08:19:22.658345Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO 09-17 04:19:22 awq_marlin.py:93] Detected that the model can run with awq_marlin, however you specified quantization=awq explicitly, so forcing awq. Use quantization=awq_marlin for faster inference\n",
"WARNING 09-17 04:19:22 config.py:338] awq quantization is not fully optimized yet. The speed can be slower than non-quantized models.\n",
"INFO 09-17 04:19:22 llm_engine.py:223] Initializing an LLM engine (v0.6.1.post2) with config: model='mychen76/Llama-3.1-8B-bonito-v1-awq', speculative_config=None, tokenizer='mychen76/Llama-3.1-8B-bonito-v1-awq', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, rope_scaling=None, rope_theta=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.float16, max_seq_len=4096, download_dir=None, load_format=LoadFormat.AUTO, tensor_parallel_size=1, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=awq, enforce_eager=False, kv_cache_dtype=auto, quantization_param_path=None, device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='outlines'), observability_config=ObservabilityConfig(otlp_traces_endpoint=None, collect_model_forward_time=False, collect_model_execute_time=False), seed=0, served_model_name=mychen76/Llama-3.1-8B-bonito-v1-awq, use_v2_block_manager=False, num_scheduler_steps=1, enable_prefix_caching=False, use_async_output_proc=True)\n",
"INFO 09-17 04:19:23 model_runner.py:997] Starting to load model mychen76/Llama-3.1-8B-bonito-v1-awq...\n",
"INFO 09-17 04:19:23 weight_utils.py:242] Using model weights format ['*.safetensors']\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "0e53544149e94bacbc0a1c854690f32a",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Loading safetensors checkpoint shards: 0% Completed | 0/2 [00:00<?, ?it/s]\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO 09-17 04:19:24 model_runner.py:1008] Loading model weights took 5.3745 GB\n",
"INFO 09-17 04:19:26 gpu_executor.py:122] # GPU blocks: 6132, # CPU blocks: 2048\n",
"INFO 09-17 04:19:28 model_runner.py:1311] Capturing the model for CUDA graphs. This may lead to unexpected consequences if the model is not static. To run the model in eager mode, set 'enforce_eager=True' or use '--enforce-eager' in the CLI.\n",
"INFO 09-17 04:19:28 model_runner.py:1315] CUDA graphs can take additional 1~3 GiB memory per GPU. If you are running out of memory, consider decreasing `gpu_memory_utilization` or enforcing eager mode. You can also reduce the `max_num_seqs` as needed to decrease memory usage.\n",
"INFO 09-17 04:19:39 model_runner.py:1430] Graph capturing finished in 11 secs.\n"
]
}
],
"source": [
"from vllm import LLM, SamplingParams\n",
"import torch \n",
"\n",
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
"model_id = 'mychen76/Llama-3.1-8B-bonito-v1-awq'\n",
"\n",
"# Create an LLM.\n",
"vbonito = VLLMBonito(model=model_id, \n",
" quantization=\"AWQ\",\n",
" gpu_memory_utilization=0.80,\n",
" max_seq_len_to_capture=4096,\n",
" max_model_len=4096,\n",
" speculative_max_model_len = 4096)\n"
]
},
{
"cell_type": "markdown",
"id": "e04ccba5-907b-406f-8396-f880d371273d",
"metadata": {},
"source": [
"#### generate \"nli\": \"natural language inference\""
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "50b8baac-18c8-41bd-b126-190119087ccc",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:20:10.531375Z",
"iopub.status.busy": "2024-09-17T08:20:10.531202Z",
"iopub.status.idle": "2024-09-17T08:20:11.851687Z",
"shell.execute_reply": "2024-09-17T08:20:11.851225Z",
"shell.execute_reply.started": "2024-09-17T08:20:10.531362Z"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Processed prompts: 100%|██████████████████| 10/10 [00:01<00:00, 7.72it/s, est. speed input: 757.82 toks/s, output: 359.21 toks/s]\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "37f9beb99d0448fba399cd2cb1dad305",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Filter: 0%| | 0/10 [00:00<?, ? examples/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "90458d9887d04da3b2b2bc5d101e6fee",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Map: 0%| | 0/10 [00:00<?, ? examples/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"'----Generated Instructions----'\n",
"(\"Input: ['Given that 2.3 Provided that the Recipient has a written agreement \"\n",
" 'with the following persons or entities requiring them to treat the '\n",
" 'Confidential Information in accordance with this Agreement, the Recipient '\n",
" 'may disclose the Confidential Information to: 2.3.1 Any other party with '\n",
" 'the Discloser’s prior written consent; and 2.3.2 the Recipient’s employees, '\n",
" 'officials, representatives and agents who have a strict need to know the '\n",
" 'contents of the Confidential Information, and employees, officials, '\n",
" 'representatives and agents of any legal entity that it controls, controls '\n",
" 'it, or with which it is under common control, who have a similar need to '\n",
" 'know the contents of the Confidential Information, provided that, for these '\n",
" 'purposes a controlled legal entity means: Does it follow that The Recipient '\n",
" \"is a person Yes, no, or maybe?', '5. All Confidential Information in any \"\n",
" 'form and any medium, including all copies thereof, disclosed to the '\n",
" 'Recipient shall be returned to UNHCR or destroyed: (a) if a business '\n",
" 'relationship is not entered into with UNHCR on or before the date which is '\n",
" 'three (3) months after the date both Parties have signed the Agreement; or '\n",
" 'Based on that information, is the claim: \"The UNHCR is not a recipient.\" '\n",
" \"true, false, or inconclusive?', 'Given 4. Nothing in this Agreement is to be \"\n",
" 'construed as granting the Recipient, by implication or otherwise, any right '\n",
" 'whatsoever with respect to the Confidential Information or part thereof. Is '\n",
" 'it guaranteed true that \"The Recipient is not granted any rights to the '\n",
" 'Confidential Information.\"? Yes, no, or maybe?\\', \\'11. The Recipient shall '\n",
" 'not advertise or otherwise make public the fact that it has a confidential '\n",
" 'relationship with UNHCR, nor shall the Recipient, in any manner whatsoever '\n",
" 'use the name, emblem, or official seal of the United Nations or UNHCR, or '\n",
" 'any abbreviation of the name of the United Nations or UNHCR in connection '\n",
" 'with its business or otherwise. \\\\n\\\\nQuestion: Does this imply that \"The '\n",
" 'United Nations is a good organization.\"? Yes, no, or maybe?\\', \\'1. '\n",
" '“Confidential Information”, whenever used in this Agreement, shall mean any '\n",
" 'data, document, specification and other information or material, that is '\n",
" 'delivered or disclosed by UNHCR to the Recipient in any form whatsoever, '\n",
" 'whether orally, visually in writing or otherwise (including computerized '\n",
" 'form), and that, at the time of disclosure to the Recipient, is designated '\n",
" 'as confidential.\\\\nQuestion: The UNHCR is a non-governmental organization. '\n",
" \"True, False, or Neither?', '1. “Confidential Information”, whenever used in \"\n",
" 'this Agreement, shall mean any data, document, specification and other '\n",
" 'information or material, that is delivered or disclosed by UNHCR to the '\n",
" 'Recipient in any form whatsoever, whether orally, visually in writing or '\n",
" 'otherwise (including computerized form), and that, at the time of disclosure '\n",
" 'to the Recipient, is designated as confidential. Based on that information, '\n",
" 'is the claim: \"Confidential Information is only delivered to the Recipient '\n",
" 'in written form.\" true, false, or inconclusive?\\', \\'Either Party may '\n",
" 'terminate the working relationship contemplated by this Agreement by '\n",
" 'providing written notice to the other, provided, however, that the '\n",
" 'obligations and restrictions hereunder regarding the Confidential '\n",
" 'Information shall remain effective following any such termination or any '\n",
" 'other termination or expiration of this Agreement. Using only the above '\n",
" 'description and the fact that \"The obligations and restrictions hereunder '\n",
" 'regarding the Confidential Information shall remain effective following any '\n",
" 'such termination or any other termination or expiration of this Agreement.\" '\n",
" 'is true, is \"The obligations and restrictions hereunder regarding the '\n",
" 'Confidential Information shall remain effective following any such '\n",
" 'termination or any other termination or expiration of this Agreement. It is '\n",
" 'a very long sentence.\" always, sometimes, or never true?\\', \\'The Recipient '\n",
" 'shall not be precluded from disclosing the Confidential Information that is '\n",
" '(i) obtained by the Recipient without restriction from a third party who is '\n",
" 'not in breach of any obligation as to confidentiality to the owner of such '\n",
" 'Confidential Information or any other person, or (ii) disclosed by the '\n",
" 'Discloser to a third party without any obligation of confidentiality, or '\n",
" '(iii) previously known by the Recipient, or (iv) at any time is developed '\n",
" 'by the Recipient completely independently of any disclosures hereunder. '\n",
" '\\\\n\\\\nQuestion: Does this imply that \"Confidential Information has been seen '\n",
" 'by carl\"? Yes, no, or maybe?\\', \\'Assume it is true that The Recipient shall '\n",
" 'not be precluded from disclosing the Confidential Information that is (i) '\n",
" 'obtained by the Recipient without restriction from a third party who is not '\n",
" 'in breach of any obligation as to confidentiality to the owner of such '\n",
" 'Confidential Information or any other person, or \\\\n\\\\nTherefore, \"The '\n",
" 'Recipient shall not be precluded from disclosing the Confidential '\n",
" 'Information that is obtained by the Recipient without restriction from a '\n",
" 'third party who is not in breach of any obligation as to confidentiality to '\n",
" 'the owner of such Confidential Information or any other person, or the owner '\n",
" 'of such Confidential Information.\" is guaranteed, possible, or '\n",
" \"impossible?', '1. “Confidential Information”, whenever used in this \"\n",
" 'Agreement, shall mean any data, document, specification and other '\n",
" 'information or material, that is delivered or disclosed by UNHCR to the '\n",
" 'Recipient in any form whatsoever, whether orally, visually in writing or '\n",
" 'otherwise (including computerized form), and that, at the time of disclosure '\n",
" 'to the Recipient, is designated as confidential. \\\\n\\\\nKeeping in mind the '\n",
" 'above text, consider: Confidential Information is not protected by the '\n",
" \"agreement. Is this always, sometimes, or never correct?']\")\n",
"(\"Output: ['Maybe', 'False', 'Yes', 'Maybe', 'Neither', 'False', 'Sometimes', \"\n",
" \"'Maybe', 'Guaranteed', 'Never']\")\n"
]
}
],
"source": [
"from vllm import SamplingParams\n",
"from transformers import set_seed\n",
"from pprint import pprint\n",
"set_seed(2)\n",
"\n",
"# Generate synthetic instruction tuning dataset\n",
"sampling_params = SamplingParams(max_tokens=256, top_p=0.95, temperature=0.5, n=1)\n",
"\n",
"synthetic_dataset = vbonito.generate_tasks(\n",
" unannotated_text_ds,\n",
" context_col=\"input\",\n",
" task_type=\"nli\",\n",
" sampling_params=sampling_params\n",
")\n",
"\n",
"pprint(\"----Generated Instructions----\")\n",
"pprint(f'Input: {synthetic_dataset[\"input\"]}')\n",
"pprint(f'Output: {synthetic_dataset[\"output\"]}')"
]
},
{
"cell_type": "markdown",
"id": "ddedcdb4-6d2c-4fa0-8508-3521285d460d",
"metadata": {},
"source": [
"#### \"exqa\": \"extractive question answering\","
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "38b09b41-d044-48f4-aabc-16634e8e99c4",
"metadata": {
"execution": {
"iopub.execute_input": "2024-09-17T08:24:37.023979Z",
"iopub.status.busy": "2024-09-17T08:24:37.023731Z",
"iopub.status.idle": "2024-09-17T08:24:38.025746Z",
"shell.execute_reply": "2024-09-17T08:24:38.025384Z",
"shell.execute_reply.started": "2024-09-17T08:24:37.023955Z"
}
},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "5a59db3acc764c03b00c3a683f5c2827",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Map: 0%| | 0/10 [00:00<?, ? examples/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"Processed prompts: 100%|█████████████████| 10/10 [00:00<00:00, 10.39it/s, est. speed input: 1030.27 toks/s, output: 429.36 toks/s]\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "1b8deb9aa5564b498249d91d6534e912",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Filter: 0%| | 0/10 [00:00<?, ? examples/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "489786c4cc6f4a67bbc2f2df0fe36506",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Map: 0%| | 0/10 [00:00<?, ? examples/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"'----Generated Instructions----'\n",
"('Input: [\\'Given the following passage\\\\n\\\\n\"2.3 Provided that the Recipient '\n",
" 'has a written agreement with the following persons or entities requiring '\n",
" 'them to treat the Confidential Information in accordance with this '\n",
" 'Agreement, the Recipient may disclose the Confidential Information to: '\n",
" '2.3.1 Any other party with the Discloser’s prior written consent; and 2.3.2 '\n",
" 'the Recipient’s employees, officials, representatives and agents who have a '\n",
" 'strict need to know the contents of the Confidential Information, and '\n",
" 'employees, officials, representatives and agents of any legal entity that it '\n",
" 'controls, controls it, or with which it is under common control, who have a '\n",
" 'similar need to know the contents of the Confidential Information, provided '\n",
" 'that, for these purposes a controlled legal entity means:\",\\\\n\\\\nanswer the '\n",
" 'following question: What is the third requirement for the Recipient to '\n",
" \"disclose the Confidential Information?', '5. All Confidential Information in \"\n",
" 'any form and any medium, including all copies thereof, disclosed to the '\n",
" 'Recipient shall be returned to UNHCR or destroyed: (a) if a business '\n",
" 'relationship is not entered into with UNHCR on or before the date which is '\n",
" 'three (3) months after the date both Parties have signed the Agreement; '\n",
" 'or\\\\n\\\\nQ: What is the maximum time allowed for a business relationship to '\n",
" 'be entered into?\\\\n\\\\nA:\\', \\'Question: \"What is the name of the person that '\n",
" 'is receiving information?\"\\\\n\\\\nContext: \"4. Nothing in this Agreement is to '\n",
" 'be construed as granting the Recipient, by implication or otherwise, any '\n",
" 'right whatsoever with respect to the Confidential Information or part '\n",
" 'thereof.\"\\\\n\\\\nAnswer:\\', \\'Refer to the passage below and answer the '\n",
" 'following question:\\\\n\\\\nPassage: 11. The Recipient shall not advertise or '\n",
" 'otherwise make public the fact that it has a confidential relationship with '\n",
" 'UNHCR, nor shall the Recipient, in any manner whatsoever use the name, '\n",
" 'emblem, or official seal of the United Nations or UNHCR, or any abbreviation '\n",
" 'of the name of the United Nations or UNHCR in connection with its business '\n",
" 'or otherwise.\\\\n\\\\nQuestion: What is the Recipient prohibited from doing '\n",
" \"with the name, emblem, or official seal of the United Nations?', '1. \"\n",
" '“Confidential Information”, whenever used in this Agreement, shall mean any '\n",
" 'data, document, specification and other information or material, that is '\n",
" 'delivered or disclosed by UNHCR to the Recipient in any form whatsoever, '\n",
" 'whether orally, visually in writing or otherwise (including computerized '\n",
" 'form), and that, at the time of disclosure to the Recipient, is designated '\n",
" 'as confidential.\\\\n\\\\nQ: What is the term for information that is delivered '\n",
" \"or disclosed by UNHCR to the Recipient?\\\\n\\\\nA:', '1. “Confidential \"\n",
" 'Information”, whenever used in this Agreement, shall mean any data, '\n",
" 'document, specification and other information or material, that is delivered '\n",
" 'or disclosed by UNHCR to the Recipient in any form whatsoever, whether '\n",
" 'orally, visually in writing or otherwise (including computerized form), and '\n",
" 'that, at the time of disclosure to the Recipient, is designated as '\n",
" 'confidential.\\\\n\\\\nQ: What is the term for information that is given to the '\n",
" 'Recipient?\\\\n\\\\nReferring to the passage above, the correct answer to the '\n",
" \"question is', 'Either Party may terminate the working relationship \"\n",
" 'contemplated by this Agreement by providing written notice to the other, '\n",
" 'provided, however, that the obligations and restrictions hereunder regarding '\n",
" 'the Confidential Information shall remain effective following any such '\n",
" 'termination or any other termination or expiration of this '\n",
" 'Agreement.\\\\n\\\\nWith reference to the above context, What is the only way '\n",
" \"the working relationship can be terminated?', 'The Recipient shall not be \"\n",
" 'precluded from disclosing the Confidential Information that is (i) obtained '\n",
" 'by the Recipient without restriction from a third party who is not in breach '\n",
" 'of any obligation as to confidentiality to the owner of such Confidential '\n",
" 'Information or any other person, or (ii) disclosed by the Discloser to a '\n",
" 'third party without any obligation of confidentiality, or (iii) previously '\n",
" 'known by the Recipient, or (iv) at any time is developed by the Recipient '\n",
" 'completely independently of any disclosures hereunder.\\\\n\\\\nQ: Who can the '\n",
" 'Recipient disclose confidential information to without breaching the '\n",
" \"agreement?\\\\n\\\\nA:', 'The Recipient shall not be precluded from disclosing \"\n",
" 'the Confidential Information that is (i) obtained by the Recipient without '\n",
" 'restriction from a third party who is not in breach of any obligation as to '\n",
" 'confidentiality to the owner of such Confidential Information or any other '\n",
" 'person, or\\\\n\\\\nQ: What is a third party not in breach of any obligation as '\n",
" 'to confidentiality to the owner of such Confidential Information or any '\n",
" 'other person?\\\\n\\\\nReferring to the passage above, the correct answer to the '\n",
" \"question is', '1. “Confidential Information”, whenever used in this \"\n",
" 'Agreement, shall mean any data, document, specification and other '\n",
" 'information or material, that is delivered or disclosed by UNHCR to the '\n",
" 'Recipient in any form whatsoever, whether orally, visually in writing or '\n",
" 'otherwise (including computerized form), and that, at the time of disclosure '\n",
" 'to the Recipient, is designated as confidential.\\\\n\\\\nWith reference to the '\n",
" \"above context, What is Confidential Information?']\")\n",
"(\"Output: ['a strict need to know', 'three (3) months', 'Recipient', 'use', \"\n",
" \"'Confidential Information', 'Confidential Information', 'written notice to \"\n",
" \"the other', 'a third party who is not in breach of any obligation as to \"\n",
" \"confidentiality', 'obtained by the Recipient without restriction', 'any \"\n",
" 'data, document, specification and other information or material, that is '\n",
" \"delivered or disclosed by UNHCR to the Recipient']\")\n"
]
}
],
"source": [
"from vllm import SamplingParams\n",
"from transformers import set_seed\n",
"from pprint import pprint\n",
"set_seed(2)\n",
"\n",
"# Generate synthetic instruction tuning dataset\n",
"sampling_params = SamplingParams(max_tokens=256, top_p=0.95, temperature=0.5, n=1)\n",
"\n",
"synthetic_dataset = vbonito.generate_tasks(\n",
" unannotated_text_ds,\n",
" context_col=\"input\",\n",
" task_type=\"exqa\",\n",
" sampling_params=sampling_params\n",
")\n",
"\n",
"pprint(\"----Generated Instructions----\")\n",
"pprint(f'Input: {synthetic_dataset[\"input\"]}')\n",
"pprint(f'Output: {synthetic_dataset[\"output\"]}')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b3b45cc5-e863-413e-939f-ac00b2c2565a",
"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": "3.10.12"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|