File size: 25,441 Bytes
4d5aebb 46068d3 4d5aebb |
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 |
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "f41486ad",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"NVIDIA A100-PCIE-40GB\n"
]
}
],
"source": [
"# step 0. Preliminary\n",
"import torch\n",
"# check that cuda doesn't crash on us\n",
"print(torch.cuda.get_device_name())\n",
"# check that transformers installed\n",
"import transformers"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "ffd19cfb",
"metadata": {},
"outputs": [],
"source": [
"EPOCHS=3"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "3a91ef1f",
"metadata": {},
"outputs": [],
"source": [
"# Step 1. Preparing the training\n",
"# First ensure that required files are here\n",
"from pathlib import Path\n",
"assert Path(\"TinyStoriesV2-GPT4-train.txt\").exists()\n",
"assert Path(\"TinyStoriesV2-GPT4-valid.txt\").exists()"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "56b046d5",
"metadata": {},
"outputs": [],
"source": [
"# Then prepare directories\n",
"Path(\"chunks.txt/train\").mkdir(parents=True, exist_ok=True)\n",
"Path(\"chunks.tensors/train\").mkdir(parents=True, exist_ok=True)\n",
"Path(\"chunks.txt/valid\").mkdir(parents=True, exist_ok=True)\n",
"Path(\"chunks.tensors/valid\").mkdir(parents=True, exist_ok=True)"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "1bddb2ee",
"metadata": {},
"outputs": [],
"source": [
"# Then prepare method to split one text to several\n",
"from multiprocessing.pool import Pool\n",
"from tqdm.contrib.concurrent import process_map\n",
"import os\n",
"_chunk_me = None\n",
"def extract_chunk(chunk):\n",
" split, i, chunk_from, chunk_to = chunk\n",
" chunk = _chunk_me[chunk_from:chunk_to].strip() \n",
" name = f\"chunks.txt/{split}/chunk-{i+1}.txt\"\n",
" with open(name, \"w\") as f:\n",
" f.write(chunk)\n",
" return name\n",
"\n",
"def split_to_text_chunks(split:str, chunk_size = 16*1024*1024, max_workers=None):\n",
" global _chunk_me #text is too chunky to pass as argument. storing as global so fork() can take care of it\n",
" print(f\"reading {split}\")\n",
" text = _chunk_me = Path(f\"./TinyStoriesV2-GPT4-{split}.txt\").read_text()\n",
" offsets = [] \n",
" delimiter = \"<|endoftext|>\"\n",
" i=0\n",
" while i < len(text): \n",
" offsets.append(i)\n",
" i += chunk_size\n",
" i = text.find(delimiter, i)\n",
" if i < 0:\n",
" break\n",
" i += len(delimiter)\n",
" offsets.append(len(text))\n",
" chunks = [(split, i, start,end) for (i, (start, end)) in enumerate(zip(offsets[:-1], offsets[1:]))]\n",
" \n",
" print(\"writing\")\n",
" process_map(extract_chunk, chunks, max_workers=max_workers)\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "e60017ee",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Assuming split has finished already\n"
]
}
],
"source": [
"# Prepare text of train split\n",
"if not Path(\"chunks.txt/train/chunk-133.txt\").exists():\n",
" split_to_text_chunks(\"train\")\n",
"else:\n",
" print(\"Assuming split has finished already\")"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "e9b7effe",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Assuming split has finished already\n"
]
}
],
"source": [
"# Prepare text of valid split\n",
"if not Path(\"chunks.txt/valid/chunk-2.txt\").exists():\n",
" split_to_text_chunks(\"valid\") \n",
"else:\n",
" print(\"Assuming split has finished already\")"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "b4706f24",
"metadata": {},
"outputs": [],
"source": [
"# Step 2. Prepare OpenLLAMA tokenizer. \n",
"#Needed to be done once(TODO: add code to load tokenizer?)\n",
"from transformers import AutoTokenizer\n",
"import os\n",
"if not Path('tokenizer.json').exists(): \n",
" try:\n",
" tokenizer = AutoTokenizer.from_pretrained(\"openlm-research/open_llama_3b\")\n",
" tokenizer.save_pretrained(\".\")\n",
" except Exception as e:\n",
" print(e)\n",
" os.environ[\"PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION\"]=\"python\" \n",
" tokenizer = AutoTokenizer.from_pretrained(\"openlm-research/open_llama_3b\")\n",
" tokenizer.save_pretrained(\".\")\n",
" del os.environ[\"PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION\"]\n",
"tokenizer = AutoTokenizer.from_pretrained(\".\")"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "f9c935b0",
"metadata": {},
"outputs": [],
"source": [
"# Step 3. Preparing to tokenize each text chunk\n",
"from tqdm.contrib.concurrent import process_map\n",
"def tokenize_file(filename:Path):\n",
" text = Path.read_text(filename)\n",
" stories = text.split(\"<|endoftext|>\")\n",
" result = []\n",
" while stories:\n",
" story = stories.pop(0).strip()\n",
" tokenized = tokenizer(story, max_length=None).input_ids\n",
" tokenized.append(tokenizer.eos_token_id)\n",
" result.append(torch.tensor(tokenized))\n",
" output_name = str(filename).replace(\".txt\", \".tensors\")\n",
" torch.save(result, output_name)\n",
"\n",
"def tokenize_split(split, max_workers=None):\n",
" to_process = list(Path(f\"chunks.txt/{split}\").glob(\"*\")) \n",
" process_map(tokenize_file, to_process, max_workers=max_workers)\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "95257f12",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Assuming train was tokenized already\n"
]
}
],
"source": [
"# processing train(this can take several minutes)\n",
"if not Path(\"chunks.tensors/train/chunk-133.tensors\").exists():\n",
" tokenize_split(\"train\")\n",
"else:\n",
" print(\"Assuming train was tokenized already\")"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "bbbe4599",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Assuming valid was tokenized already\n"
]
}
],
"source": [
"# processing valid(this can take one minutes)\n",
"if not Path(\"chunks.tensors/valid/chunk-2.tensors\").exists():\n",
" tokenize_split(\"valid\")\n",
"else:\n",
" print(\"Assuming valid was tokenized already\")"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "a31a4aa7",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Resetting [PAD] to [EOS]\n"
]
}
],
"source": [
"# Step 4. Training. \n",
"# Step 4.1 Preparing tokenizer and setting pad token if it is not set(it is not set)\n",
"tokenizer = AutoTokenizer.from_pretrained(\".\")\n",
"if not tokenizer.pad_token_id:\n",
" tokenizer.pad_token_id = tokenizer.eos_token_id\n",
" print(\"Resetting [PAD] to [EOS]\")"
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "f677c9c0",
"metadata": {
"scrolled": true
},
"outputs": [],
"source": [
"# Step 4.2. Preparing model\n",
"from transformers.models.llama.modeling_llama import LlamaConfig, LlamaForCausalLM\n",
"\n",
"tiny_llama = LlamaConfig(\n",
" hidden_size=64, \n",
" vocab_size=tokenizer.vocab_size,\n",
" intermediate_size=256, \n",
" num_attention_heads=16, \n",
" num_hidden_layers=8)\n",
"\n",
"torch.manual_seed(11010)\n",
"torch.cuda.manual_seed(11010)\n",
"model = LlamaForCausalLM(tiny_llama).cuda().bfloat16()"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "aad9620b",
"metadata": {},
"outputs": [],
"source": [
"import functools\n",
"import torch.nn.functional as F\n",
"from tqdm.contrib.concurrent import process_map\n",
"from tqdm.auto import tqdm\n",
"\n",
"# Step 4.3 Preparing dataset class\n",
"def get_file_data_len(filename):\n",
" data = torch.load(filename)\n",
" return (filename, len(data))\n",
"from datasets import Dataset\n",
"\n",
"CACHE_SIZE = 2000 # There are ~150 train splits. We can fit them in memory, so let's do it\n",
"\n",
"class TinyDataset(Dataset):\n",
" def __init__(self, split: str, populate_cache=True):\n",
" print(f\"Reading dataset {split} data\")\n",
" self.file_lens = process_map(\n",
" get_file_data_len,\n",
" list(Path(f\"chunks.tensors/{split}\").glob(\"*\")))\n",
" self.file_lens.sort()\n",
" if populate_cache:\n",
" print(\"Populating a cache\")\n",
" for filename, _ in tqdm(self.file_lens):\n",
" self.load_tensor_file(filename)\n",
"\n",
" @functools.lru_cache(maxsize=CACHE_SIZE)\n",
" def load_tensor_file(self, filename):\n",
" return torch.load(filename)\n",
"\n",
" def __len__(self):\n",
" return sum(x[1] for x in self.file_lens)\n",
"\n",
" def global_index_to_local(self, i):\n",
" for (file, length) in self.file_lens:\n",
" if i < length:\n",
" return (file, i)\n",
" i -= length\n",
" raise IndexError(f\"{i} is out-of-bonds, have {len(self)} sample\")\n",
"\n",
" def __getitem__(self, index):\n",
" if torch.is_tensor(index):\n",
" index = index.tolist()\n",
" if isinstance(index, int):\n",
" filename, local_index = self.global_index_to_local(index)\n",
" tensors = self.load_tensor_file(filename)\n",
" return {\n",
" 'input_ids': tensors[local_index]\n",
" }\n",
" if isinstance(index, list):\n",
" data = []\n",
" indices = index\n",
" for index in indices:\n",
" filename, local_index = self.global_index_to_local(index)\n",
" tensors = self.load_tensor_file(filename)\n",
" data.append(tensors[local_index])\n",
"\n",
" return {'input_ids': data}\n",
"\n",
" raise TypeError(f'Invaldi index type {type(index)}')\n",
" \n",
"def batch_collate(data: list[torch.Tensor]):\n",
" max_len = max(len(datum[\"input_ids\"]) for datum in data)\n",
" inputs = []\n",
" attentions = []\n",
" for row in data:\n",
" input_ids = row[\"input_ids\"]\n",
" attention_mask = torch.ones_like(input_ids)\n",
" attention_mask[-1] = 0 # don't care about EOS\n",
" # Manual padding\n",
" to_pad = max_len - len(input_ids)\n",
" is_left_pad = tokenizer.padding_side == \"left\"\n",
" padding = (is_left_pad * to_pad, (1 - is_left_pad) * to_pad)\n",
" input_ids = F.pad(input_ids, padding, value=tokenizer.pad_token_id)\n",
" attention_mask = F.pad(attention_mask, padding, value=0)\n",
" inputs.append(input_ids)\n",
" attentions.append(attention_mask)\n",
"\n",
" attention_masks = torch.stack(attentions)\n",
" input_ids = torch.stack(inputs)\n",
" labels = input_ids.clone()\n",
"\n",
" # disable prediction of the padding\n",
" labels[attention_masks == 0] = -100\n",
" # enable prediction of an actual EOS\n",
" labels[:, -1] = tokenizer.eos_token_id\n",
"\n",
" return {\n",
" 'input_ids': input_ids,\n",
" 'attention_mask': attention_masks,\n",
" 'labels': labels\n",
" }\n",
"\n",
"def get_max_story_length(ds): \n",
" return max(file_len[1] for file_len in ds.file_lens)\n"
]
},
{
"cell_type": "code",
"execution_count": 17,
"id": "2e828afe",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Reading dataset train data\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "8ca542afc1694073af6dcf9ce5f7e13a",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/133 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Populating a cache\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "8035a75107e84a54870a8c6f15c4100a",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/133 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"ename": "AssertionError",
"evalue": "WARNIING: split long stories",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mAssertionError\u001b[0m Traceback (most recent call last)",
"Cell \u001b[0;32mIn[17], line 3\u001b[0m\n\u001b[1;32m 1\u001b[0m \u001b[38;5;28;01massert\u001b[39;00m tokenizer\u001b[38;5;241m.\u001b[39mpadding_side \u001b[38;5;129;01min\u001b[39;00m [\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mleft\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mright\u001b[39m\u001b[38;5;124m\"\u001b[39m]\n\u001b[1;32m 2\u001b[0m train_ds \u001b[38;5;241m=\u001b[39m TinyDataset(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtrain\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m----> 3\u001b[0m \u001b[38;5;28;01massert\u001b[39;00m get_max_story_length(train_ds) \u001b[38;5;241m<\u001b[39m\u001b[38;5;241m=\u001b[39m tokenizer\u001b[38;5;241m.\u001b[39mmodel_max_length, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mWARNIING: split long stories\u001b[39m\u001b[38;5;124m\"\u001b[39m\n",
"\u001b[0;31mAssertionError\u001b[0m: WARNIING: split long stories"
]
}
],
"source": [
"assert tokenizer.padding_side in [\"left\", \"right\"]\n",
"train_ds = TinyDataset(\"train\")\n",
"assert get_max_story_length(train_ds) <= tokenizer.model_max_length, \"WARNIING: split long stories\""
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "6412e7c5",
"metadata": {},
"outputs": [],
"source": [
"from torch.utils.data import DataLoader\n",
"torch.manual_seed(11010)\n",
"torch.cuda.manual_seed(11010)\n",
"train_dl = DataLoader(train_ds, 16, True, collate_fn=batch_collate)"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "f3ff5a66",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mggg4\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n"
]
},
{
"data": {
"text/html": [
"Tracking run with wandb version 0.15.5"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"Run data is saved locally in <code>/home/mayk/tiny-llama/wandb/run-20230707_181234-rilt4m6f</code>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"Syncing run <strong><a href='https://wandb.ai/ggg4/training-tiny-llama-preview/runs/rilt4m6f' target=\"_blank\">grateful-jazz-4</a></strong> to <a href='https://wandb.ai/ggg4/training-tiny-llama-preview' target=\"_blank\">Weights & Biases</a> (<a href='https://wandb.me/run' target=\"_blank\">docs</a>)<br/>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" View project at <a href='https://wandb.ai/ggg4/training-tiny-llama-preview' target=\"_blank\">https://wandb.ai/ggg4/training-tiny-llama-preview</a>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" View run at <a href='https://wandb.ai/ggg4/training-tiny-llama-preview/runs/rilt4m6f' target=\"_blank\">https://wandb.ai/ggg4/training-tiny-llama-preview/runs/rilt4m6f</a>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<button onClick=\"this.nextSibling.style.display='block';this.style.display='none';\">Display W&B run</button><iframe src='https://wandb.ai/ggg4/training-tiny-llama-preview/runs/rilt4m6f?jupyter=true' style='border:none;width:100%;height:420px;display:none;'></iframe>"
],
"text/plain": [
"<wandb.sdk.wandb_run.Run at 0x7f6af8170b50>"
]
},
"execution_count": 20,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# prepare wandb\n",
"import wandb\n",
"wandb.init(\n",
" project=\"training-tiny-llama-preview\",\n",
" config={\n",
" \"architecture\": \"llama\",\n",
" \"dataset\": \"tiny-stories\",\n",
" \"epochs\": EPOCHS,\n",
" } \n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "aed7b7a4",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 21,
"id": "166a4a27",
"metadata": {},
"outputs": [],
"source": [
"from tqdm.auto import tqdm\n",
"def save_imm(epoch, step, saved=[]):\n",
" fname = f\"step-{epoch}-{step}.bin\"\n",
" torch.save(model.state_dict(), f\"step-{epoch}-{step}.bin\")\n",
" saved.append(fname)\n",
" if len(saved) > 5:\n",
" delete_me = saved.pop(0)\n",
" Path(delete_me).unlink(missing_ok=True)\n",
"\n",
"def epoch_step(epoch, opt):\n",
" for i, batch in enumerate(bar := tqdm(train_dl)):\n",
" for k in batch:\n",
" batch[k] = batch[k].to(device=model.lm_head.weight.device)\n",
" \n",
" n_batch, n_seq = batch[\"input_ids\"].shape\n",
" if n_seq > tokenizer.model_max_length:\n",
" assert tokenizer.padding_side == \"right\", \"Left-pad truncation only supported[as model should not see >2k token anyway]\"\n",
" batch[\"input_ids\"] = batch[\"input_ids\"][:, -tokenizer.model_max_length]\n",
" batch[\"labels\"] = batch[\"labels\"][:, -tokenizer.model_max_length]\n",
" batch[\"attention_mask\"] = batch[\"attention_mask\"][:, -tokenizer.model_max_length]\n",
" \n",
" \n",
" loss = model(**batch).loss\n",
" loss.backward()\n",
" opt.step()\n",
" opt.zero_grad()\n",
" bar.set_description(f'L:{loss.item():.4f}')\n",
" wandb.log({\"loss\": loss.item()})\n",
" if (i+1) % 100 == 0:\n",
" save_imm(epoch, i+1)\n",
" \n",
" torch.save(model.state_dict(), f\"epoch-{epoch}.bin\")\n"
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "ec4943c7",
"metadata": {},
"outputs": [],
"source": [
"opt = torch.optim.AdamW(model.parameters(), fused=True)\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "daae9020",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "f7ab6fe3b99546f49acb0d43888b7ceb",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/169865 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"for e in range(EPOCHS):\n",
" epoch_step(e+1, opt)"
]
},
{
"cell_type": "code",
"execution_count": 45,
"id": "87988cf5",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" total used free shared buff/cache available\r\n",
"Mem: 85Gi 1.5Gi 72Gi 8.0Mi 11Gi 83Gi\r\n",
"Swap: 0B 0B 0B\r\n"
]
}
],
"source": [
"!free -h"
]
},
{
"cell_type": "code",
"execution_count": 65,
"id": "e43eb9f3",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Fri Jul 7 17:44:05 2023 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 520.61.05 Driver Version: 520.61.05 CUDA Version: 11.8 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 NVIDIA A100-PCI... On | 00000000:05:00.0 Off | 0 |\n",
"| N/A 30C P0 34W / 250W | 5739MiB / 40960MiB | 0% Default |\n",
"| | | Disabled |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"| 0 N/A N/A 13768 C /opt/conda/bin/python 5736MiB |\n",
"+-----------------------------------------------------------------------------+\n"
]
}
],
"source": [
"!nvidia-smi"
]
},
{
"cell_type": "code",
"execution_count": 73,
"id": "0351f57f",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"Parameter containing:\n",
"tensor([[ 8.3618e-03, 3.8330e-02, -5.9204e-03, ..., 2.0752e-02,\n",
" 4.4861e-03, 1.2512e-02],\n",
" [ 3.9978e-03, 2.1118e-02, -3.5645e-02, ..., -1.6846e-02,\n",
" 5.0659e-03, -3.8818e-02],\n",
" [-1.6928e-05, -1.2756e-02, -1.1536e-02, ..., -1.6235e-02,\n",
" 4.8218e-03, -1.4099e-02],\n",
" ...,\n",
" [-9.8267e-03, -6.8665e-03, 1.0864e-02, ..., -1.0864e-02,\n",
" -2.4170e-02, -5.6076e-04],\n",
" [-9.5749e-04, 7.3853e-03, 4.9438e-03, ..., 1.2390e-02,\n",
" -2.1606e-02, -9.2163e-03],\n",
" [ 5.1758e-02, 2.1484e-02, -1.5381e-02, ..., -2.4292e-02,\n",
" -3.4912e-02, 3.0823e-03]], device='cuda:0', dtype=torch.bfloat16,\n",
" requires_grad=True)"
]
},
"execution_count": 73,
"metadata": {},
"output_type": "execute_result"
}
],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "ace72db5",
"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": "3.10.10"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|