Entraînement de SoraForSLM terminé

README.md

@@ -0,0 +1,54 @@
+---
+library_name: transformers
+tags:
+- generated_from_trainer
+model-index:
+- name: SoraForSLM-1
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# SoraForSLM-1
+
+This model is a fine-tuned version of [](https://huggingface.co/) on an unknown dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 0.0004
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- gradient_accumulation_steps: 4
+- total_train_batch_size: 32
+- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: linear
+- num_epochs: 5
+- mixed_precision_training: Native AMP
+
+### Training results
+
+
+
+### Framework versions
+
+- Transformers 5.0.0
+- Pytorch 2.10.0+cu128
+- Datasets 4.0.0
+- Tokenizers 0.22.2

tokenizer.json

@@ -0,0 +1,2772 @@
+{
+  "version": "1.0",
+  "truncation": null,
+  "padding": null,
+  "added_tokens": [
+    {
+      "id": 0,
+      "content": "<pad>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 1,
+      "content": "</s>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 2,
+      "content": "<s>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 3,
+      "content": "<unk>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 4,
+      "content": "<maskSub>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    {
+      "id": 5,
+      "content": "Question:",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    {
+      "id": 6,
+      "content": "Réponse:",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    {
+      "id": 2628,
+      "content": "<mask>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    }
+  ],
+  "normalizer": {
+    "type": "Replace",
+    "pattern": {
+      "String": " "
+    },
+    "content": "▁"
+  },
+  "pre_tokenizer": null,
+  "post_processor": {
+    "type": "TemplateProcessing",
+    "single": [
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      }
+    ],
+    "pair": [
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "B",
+          "type_id": 1
+        }
+      }
+    ],
+    "special_tokens": {}
+  },
+  "decoder": {
+    "type": "Sequence",
+    "decoders": [
+      {
+        "type": "Replace",
+        "pattern": {
+          "String": "▁"
+        },
+        "content": " "
+      },
+      {
+        "type": "ByteFallback"
+      },
+      {
+        "type": "Fuse"
+      }
+    ]
+  },
+  "model": {
+    "type": "BPE",
+    "dropout": null,
+    "unk_token": "<unk>",
+    "continuing_subword_prefix": null,
+    "end_of_word_suffix": null,
+    "fuse_unk": true,
+    "byte_fallback": true,
+    "ignore_merges": false,
+    "vocab": {
+      "<pad>": 0,
+      "</s>": 1,
+      "<s>": 2,
+      "<unk>": 3,
+      "<maskSub>": 4,
+      "Question:": 5,
+      "Réponse:": 6,
+      "▁the": 7,
+      "▁and": 8,
+      "▁to": 9,
+      "▁of": 10,
+      "▁a": 11,
+      "▁your": 12,
+      "▁is": 13,
+      "▁model": 14,
+      "▁data": 15,
+      "▁you": 16,
+      "▁in": 17,
+      "▁training": 18,
+      "▁with": 19,
+      "▁or": 20,
+      "▁for": 21,
+      "▁that": 22,
+      "▁are": 23,
+      "▁can": 24,
+      "▁on": 25,
+      "▁models": 26,
+      "▁by": 27,
+      "▁AI": 28,
+      "▁be": 29,
+      "▁The": 30,
+      "▁an": 31,
+      "▁it": 32,
+      "▁LLM": 33,
+      "▁as": 34,
+      "▁from": 35,
+      "▁use": 36,
+      "▁This": 37,
+      "▁language": 38,
+      "▁into": 39,
+      "▁train": 40,
+      "▁You": 41,
+      "▁may": 42,
+      "▁own": 43,
+      "▁Model": 44,
+      "▁like": 45,
+      "▁LLMs": 46,
+      "▁have": 47,
+      "▁this": 48,
+      "▁text": 49,
+      "▁Data": 50,
+      "▁learning": 51,
+      "▁will": 52,
+      "▁data.": 53,
+      "▁large": 54,
+      "▁through": 55,
+      "▁its": 56,
+      "▁more": 57,
+      "▁need": 58,
+      "▁which": 59,
+      "▁Your": 60,
+      "▁different": 61,
+      "▁even": 62,
+      "▁our": 63,
+      "▁reasoning": 64,
+      "▁using": 65,
+      "▁Language": 66,
+      "▁Large": 67,
+      "▁Training": 68,
+      "▁existing": 69,
+      "▁not": 70,
+      "▁specific": 71,
+      "▁such": 72,
+      "▁trained": 73,
+      "▁-": 74,
+      "▁how": 75,
+      "▁improve": 76,
+      "▁process": 77,
+      "▁In": 78,
+      "▁It": 79,
+      "▁about": 80,
+      "▁generate": 81,
+      "▁including": 82,
+      "▁step": 83,
+      "▁these": 84,
+      "▁used": 85,
+      "▁when": 86,
+      "▁#": 87,
+      "▁To": 88,
+      "▁custom": 89,
+      "▁feedback": 90,
+      "▁simple": 91,
+      "▁up": 92,
+      "▁For": 93,
+      "▁I": 94,
+      "▁We": 95,
+      "▁architecture": 96,
+      "▁company": 97,
+      "▁dataset": 98,
+      "▁learn": 99,
+      "▁models.": 100,
+      "▁performance": 101,
+      "▁require": 102,
+      "▁smaller": 103,
+      "▁tasks": 104,
+      "▁their": 105,
+      "▁they": 106,
+      "▁time": 107,
+      "▁access": 108,
+      "▁content": 109,
+      "▁could": 110,
+      "▁has": 111,
+      "▁if": 112,
+      "▁it’s": 113,
+      "▁make": 114,
+      "▁model,": 115,
+      "▁steps": 116,
+      "▁A": 117,
+      "▁By": 118,
+      "▁Einstein": 119,
+      "▁GPU": 120,
+      "▁Trust": 121,
+      "▁any": 122,
+      "▁data,": 123,
+      "▁do": 124,
+      "▁first": 125,
+      "▁model.": 126,
+      "▁needs": 127,
+      "▁start": 128,
+      "▁transformer": 129,
+      "▁user": 130,
+      "▁we": 131,
+      "▁you’ll": 132,
+      "▁1.": 133,
+      "▁5": 134,
+      "▁Collection": 135,
+      "▁Google": 136,
+      "▁How": 137,
+      "▁Layer": 138,
+      "▁Salesforce": 139,
+      "▁Step": 140,
+      "▁Use": 141,
+      "▁When": 142,
+      "▁all": 143,
+      "▁allows": 144,
+      "▁applications.": 145,
+      "▁create": 146,
+      "▁datasets": 147,
+      "▁don’t": 148,
+      "▁human": 149,
+      "▁most": 150,
+      "▁my": 151,
+      "▁number": 152,
+      "▁one": 153,
+      "▁output": 154,
+      "▁performance.": 155,
+      "▁power": 156,
+      "▁pre-trained": 157,
+      "▁prompt": 158,
+      "▁save": 159,
+      "▁some": 160,
+      "▁take": 161,
+      "▁Ensure": 162,
+      "▁Hugging": 163,
+      "▁These": 164,
+      "▁at": 165,
+      "▁batch": 166,
+      "▁been": 167,
+      "▁before": 168,
+      "▁better": 169,
+      "▁but": 170,
+      "▁complex": 171,
+      "▁crucial": 172,
+      "▁each": 173,
+      "▁example,": 174,
+      "▁fine-tuning": 175,
+      "▁guide": 176,
+      "▁input": 177,
+      "▁involves": 178,
+      "▁next": 179,
+      "▁other": 180,
+      "▁outputs": 181,
+      "▁parallelism": 182,
+      "▁processing": 183,
+      "▁provide": 184,
+      "▁relevant": 185,
+      "▁several": 186,
+      "▁significant": 187,
+      "▁size": 188,
+      "▁step.": 189,
+      "▁techniques": 190,
+      "▁them": 191,
+      "▁tokens": 192,
+      "▁training.": 193,
+      "▁understand": 194,
+      "▁unique": 195,
+      "▁want": 196,
+      "▁without": 197,
+      "▁words": 198,
+      "▁(e.g.": 199,
+      "▁2": 200,
+      "▁=": 201,
+      "▁API": 202,
+      "▁If": 203,
+      "▁Models": 204,
+      "▁Prompt": 205,
+      "▁They": 206,
+      "▁Train": 207,
+      "▁What": 208,
+      "▁able": 209,
+      "▁across": 210,
+      "▁apples": 211,
+      "▁articles": 212,
+      "▁between": 213,
+      "▁build": 214,
+      "▁computational": 215,
+      "▁customer": 216,
+      "▁deep": 217,
+      "▁develop": 218,
+      "▁effective": 219,
+      "▁enables": 220,
+      "▁evaluation": 221,
+      "▁get": 222,
+      "▁information": 223,
+      "▁key": 224,
+      "▁lets": 225,
+      "▁me": 226,
+      "▁might": 227,
+      "▁new": 228,
+      "▁numbers": 229,
+      "▁out": 230,
+      "▁powerful": 231,
+      "▁problems": 232,
+      "▁process,": 233,
+      "▁providers": 234,
+      "▁question": 235,
+      "▁requirements": 236,
+      "▁resources": 237,
+      "▁so": 238,
+      "▁specialized": 239,
+      "▁text.": 240,
+      "▁then": 241,
+      "▁tokenization": 242,
+      "▁tools": 243,
+      "▁training,": 244,
+      "▁transformers": 245,
+      "▁understanding": 246,
+      "▁various": 247,
+      "▁was": 248,
+      "▁web": 249,
+      "▁where": 250,
+      "▁you’re": 251,
+      "▁2.": 252,
+      "▁4.": 253,
+      "▁But": 254,
+      "▁Claude": 255,
+      "▁Claude.": 256,
+      "▁Conclusion": 257,
+      "▁Environment": 258,
+      "▁Face": 259,
+      "▁However,": 260,
+      "▁I’m": 261,
+      "▁LLM.": 262,
+      "▁LLM?": 263,
+      "▁Python": 264,
+      "▁There": 265,
+      "▁ability": 266,
+      "▁add": 267,
+      "▁al.": 268,
+      "▁amount": 269,
+      "▁analyze": 270,
+      "▁apples.": 271,
+      "▁better.": 272,
+      "▁called": 273,
+      "▁chats": 274,
+      "▁cloud": 275,
+      "▁compared": 276,
+      "▁compromising": 277,
+      "▁content,": 278,
+      "▁conversation": 279,
+      "▁conversations": 280,
+      "▁designed": 281,
+      "▁directly": 282,
+      "▁efficient": 283,
+      "▁email": 284,
+      "▁enabling": 285,
+      "▁engage": 286,
+      "▁entire": 287,
+      "▁et": 288,
+      "▁example": 289,
+      "▁fine": 290,
+      "▁further": 291,
+      "▁generic": 292,
+      "▁gives": 293,
+      "▁guide,": 294,
+      "▁handle": 295,
+      "▁human-like": 296,
+      "▁journey": 297,
+      "▁libraries": 298,
+      "▁looking": 299,
+      "▁machine": 300,
+      "▁manual": 301,
+      "▁many": 302,
+      "▁massive": 303,
+      "▁models,": 304,
+      "▁needs.": 305,
+      "▁often": 306,
+      "▁open-source": 307,
+      "▁parts": 308,
+      "▁platform": 309,
+      "▁preprocessing": 310,
+      "▁prompts": 311,
+      "▁public": 312,
+      "▁questions": 313,
+      "▁research,": 314,
+      "▁resources,": 315,
+      "▁same": 316,
+      "▁see": 317,
+      "▁sensitive": 318,
+      "▁sequences": 319,
+      "▁set": 320,
+      "▁special": 321,
+      "▁still": 322,
+      "▁support": 323,
+      "▁systems": 324,
+      "▁task": 325,
+      "▁task.": 326,
+      "▁tasks,": 327,
+      "▁think": 328,
+      "▁though": 329,
+      "▁users": 330,
+      "▁weights": 331,
+      "▁well": 332,
+      "▁we’ll": 333,
+      "▁whether": 334,
+      "▁work": 335,
+      "▁would": 336,
+      "▁(LLMs)": 337,
+      "▁(e.g.,": 338,
+      "▁(like": 339,
+      "▁10": 340,
+      "▁15,": 341,
+      "▁APIs": 342,
+      "▁Acme": 343,
+      "▁After": 344,
+      "▁Builder": 345,
+      "▁Building": 346,
+      "▁CRM": 347,
+      "▁ChatGPT": 348,
+      "▁Cloud,": 349,
+      "▁Common": 350,
+      "▁Datasets,": 351,
+      "▁Deploying": 352,
+      "▁Encoding": 353,
+      "▁Evaluating": 354,
+      "▁Figure": 355,
+      "▁FrontierMath": 356,
+      "▁GPT-4": 357,
+      "▁GPT-4,": 358,
+      "▁GPUs,": 359,
+      "▁Gemini": 360,
+      "▁Google,": 361,
+      "▁Here": 362,
+      "▁Here’s": 363,
+      "▁Install": 364,
+      "▁Instead": 365,
+      "▁Kaggle": 366,
+      "▁Keep": 367,
+      "▁LLM,": 368,
+      "▁Let's": 369,
+      "▁ML": 370,
+      "▁NLP": 371,
+      "▁Number": 372,
+      "▁Once": 373,
+      "▁OpenAI,": 374,
+      "▁Performance": 375,
+      "▁Services": 376,
+      "▁Since": 377,
+      "▁TensorFlow": 378,
+      "▁Think": 379,
+      "▁Tiers": 380,
+      "▁Tokenization": 381,
+      "▁Validation": 382,
+      "▁Web": 383,
+      "▁Whether": 384,
+      "▁Winter": 385,
+      "▁With": 386,
+      "▁You’ll": 387,
+      "▁accuracy,": 388,
+      "▁accurate": 389,
+      "▁adding": 390,
+      "▁also": 391,
+      "▁answer": 392,
+      "▁answering": 393,
+      "▁applicable": 394,
+      "▁applications": 395,
+      "▁approach": 396,
+      "▁architecture,": 397,
+      "▁around": 398,
+      "▁articles,": 399,
+      "▁artificial": 400,
+      "▁automate": 401,
+      "▁available.": 402,
+      "▁basic": 403,
+      "▁biases": 404,
+      "▁both": 405,
+      "▁bought": 406,
+      "▁capture": 407,
+      "▁cases": 408,
+      "▁chains": 409,
+      "▁choose": 410,
+      "▁code": 411,
+      "▁coding": 412,
+      "▁combine": 413,
+      "▁comes": 414,
+      "▁common": 415,
+      "▁component": 416,
+      "▁components": 417,
+      "▁computing": 418,
+      "▁concerns": 419,
+      "▁consists": 420,
+      "▁continuous": 421,
+      "▁control": 422,
+      "▁costs": 423,
+      "▁customize": 424,
+      "▁dataset.": 425,
+      "▁depending": 426,
+      "▁determine": 427,
+      "▁divided": 428,
+      "▁documents,": 429,
+      "▁does": 430,
+      "▁domains.": 431,
+      "▁down": 432,
+      "▁effectiveness": 433,
+      "▁enhance": 434,
+      "▁ensure": 435,
+      "▁ensuring": 436,
+      "▁essential": 437,
+      "▁examples": 438,
+      "▁faster": 439,
+      "▁find": 440,
+      "▁following": 441,
+      "▁framework.": 442,
+      "▁full": 443,
+      "▁function": 444,
+      "▁generated": 445,
+      "▁generative": 446,
+      "▁goal": 447,
+      "▁goes": 448,
+      "▁grounded": 449,
+      "▁grounding": 450,
+      "▁group": 451,
+      "▁help": 452,
+      "▁helps": 453,
+      "▁here": 454,
+      "▁hyperparameters": 455,
+      "▁import": 456,
+      "▁improving": 457,
+      "▁include": 458,
+      "▁includes": 459,
+      "▁infrastructure": 460,
+      "▁install": 461,
+      "▁intelligence": 462,
+      "▁language.": 463,
+      "▁latest": 464,
+      "▁lead": 465,
+      "▁level": 466,
+      "▁local": 467,
+      "▁long": 468,
+      "▁made": 469,
+      "▁means": 470,
+      "▁metrics": 471,
+      "▁model’s": 472,
+      "▁more.": 473,
+      "▁multimodal": 474,
+      "▁natural": 475,
+      "▁neural": 476,
+      "▁odd": 477,
+      "▁only": 478,
+      "▁order": 479,
+      "▁over": 480,
+      "▁part": 481,
+      "▁performance,": 482,
+      "▁performs": 483,
+      "▁popular": 484,
+      "▁possible": 485,
+      "▁potential": 486,
+      "▁prediction": 487,
+      "▁privacy": 488,
+      "▁private": 489,
+      "▁prompt.": 490,
+      "▁prompting": 491,
+      "▁range": 492,
+      "▁ready": 493,
+      "▁reinforcement": 494,
+      "▁relationships": 495,
+      "▁remain": 496,
+      "▁required": 497,
+      "▁requirements.": 498,
+      "▁research": 499,
+      "▁results": 500,
+      "▁safety": 501,
+      "▁security": 502,
+      "▁self-attention": 503,
+      "▁sequence": 504,
+      "▁services": 505,
+      "▁single": 506,
+      "▁size,": 507,
+      "▁solve": 508,
+      "▁sources": 509,
+      "▁sources,": 510,
+      "▁step-by-step": 511,
+      "▁tailored": 512,
+      "▁tasks.": 513,
+      "▁team": 514,
+      "▁technique": 515,
+      "▁text,": 516,
+      "▁than": 517,
+      "▁too": 518,
+      "▁tool": 519,
+      "▁two": 520,
+      "▁unlock": 521,
+      "▁useful": 522,
+      "▁validation": 523,
+      "▁variety": 524,
+      "▁what": 525,
+      "▁word": 526,
+      "▁\"Let's": 527,
+      "▁(2022)": 528,
+      "▁(Hint:": 529,
+      "▁(LLM)": 530,
+      "▁(MoE)": 531,
+      "▁(NLP)": 532,
+      "▁)": 533,
+      "▁1:": 534,
+      "▁1–4": 535,
+      "▁2):": 536,
+      "▁2022:": 537,
+      "▁2023:": 538,
+      "▁2:": 539,
+      "▁3.": 540,
+      "▁5.": 541,
+      "▁6": 542,
+      "▁6.": 543,
+      "▁A:": 544,
+      "▁AI,": 545,
+      "▁AI.": 546,
+      "▁APIs.": 547,
+      "▁AWS,": 548,
+      "▁Adding": 549,
+      "▁Additionally,": 550,
+      "▁Although": 551,
+      "▁An": 552,
+      "▁Anthropic.": 553,
+      "▁Architecture": 554,
+      "▁Are": 555,
+      "▁At": 556,
+      "▁BERT": 557,
+      "▁Because": 558,
+      "▁Best": 559,
+      "▁CPU,": 560,
+      "▁Choose": 561,
+      "▁Claude,": 562,
+      "▁Consider": 563,
+      "▁Continuous": 564,
+      "▁Customize": 565,
+      "▁Dataset": 566,
+      "▁Deekshitha": 567,
+      "▁Drive),": 568,
+      "▁During": 569,
+      "▁Evaluation": 570,
+      "▁Feedback": 571,
+      "▁Fine-Tuning": 572,
+      "▁First,": 573,
+      "▁GPT,": 574,
+      "▁GPT-3": 575,
+      "▁GPU,": 576,
+      "▁ID": 577,
+      "▁Imagine": 578,
+      "▁Improvement": 579,
+      "▁Incognito": 580,
+      "▁Introduction": 581,
+      "▁It’s": 582,
+      "▁I’ve": 583,
+      "▁LLM-powered": 584,
+      "▁LLM’s": 585,
+      "▁Learn": 586,
+      "▁Learning": 587,
+      "▁MCP": 588,
+      "▁Meta,": 589,
+      "▁NVIDIA).": 590,
+      "▁Now": 591,
+      "▁On": 592,
+      "▁One": 593,
+      "▁OpenAI’s": 594,
+      "▁Optimization": 595,
+      "▁Practices": 596,
+      "▁Preparation": 597,
+      "▁Prerequisites": 598,
+      "▁Privacy": 599,
+      "▁Prompt:": 600,
+      "▁Remember,": 601,
+      "▁Remove": 602,
+      "▁Researcher": 603,
+      "▁Rubin": 604,
+      "▁Salesforce’s": 605,
+      "▁Services,": 606,
+      "▁Set": 607,
+      "▁Setup": 608,
+      "▁Some": 609,
+      "▁Sree": 610,
+      "▁Stage": 611,
+      "▁Summer": 612,
+      "▁Take": 613,
+      "▁Testing": 614,
+      "▁Text": 615,
+      "▁Time:": 616,
+      "▁Trainer,": 617,
+      "▁Transformer": 618,
+      "▁Weights": 619,
+      "▁While": 620,
+      "▁Wikipedia": 621,
+      "▁Write": 622,
+      "▁Yerra": 623,
+      "▁above.": 624,
+      "▁abstracts": 625,
+      "▁academic": 626,
+      "▁accessible": 627,
+      "▁according": 628,
+      "▁accuracy": 629,
+      "▁actual": 630,
+      "▁address)": 631,
+      "▁adjusting": 632,
+      "▁advantage": 633,
+      "▁after": 634,
+      "▁against": 635,
+      "▁agent,": 636,
+      "▁allow": 637,
+      "▁allowing": 638,
+      "▁along": 639,
+      "▁always": 640,
+      "▁amounts": 641,
+      "▁analysis.": 642,
+      "▁another": 643,
+      "▁answering,": 644,
+      "▁applying": 645,
+      "▁approach.": 646,
+      "▁approaches": 647,
+      "▁appropriate": 648,
+      "▁apps.": 649,
+      "▁architecture.": 650,
+      "▁architectures": 651,
+      "▁assess": 652,
+      "▁ate": 653,
+      "▁attention": 654,
+      "▁audience": 655,
+      "▁audit": 656,
+      "▁based": 657,
+      "▁because": 658,
+      "▁behavior,": 659,
+      "▁being": 660,
+      "▁broad": 661,
+      "▁calls,": 662,
+      "▁capabilities,": 663,
+      "▁capable": 664,
+      "▁capitalization,": 665,
+      "▁case": 666,
+      "▁case.": 667,
+      "▁chain-of-thought": 668,
+      "▁challenging": 669,
+      "▁characters.": 670,
+      "▁chips,": 671,
+      "▁choose.": 672,
+      "▁chunks": 673,
+      "▁clear": 674,
+      "▁clusters": 675,
+      "▁collect": 676,
+      "▁collected": 677,
+      "▁collection": 678,
+      "▁combining": 679,
+      "▁come": 680,
+      "▁commonly": 681,
+      "▁concepts": 682,
+      "▁conduct": 683,
+      "▁configuration": 684,
+      "▁connectors": 685,
+      "▁consider": 686,
+      "▁contact": 687,
+      "▁context,”": 688,
+      "▁continue": 689,
+      "▁controls": 690,
+      "▁conversation,": 691,
+      "▁conversational": 692,
+      "▁conversations,": 693,
+      "▁copied": 694,
+      "▁core": 695,
+      "▁cover": 696,
+      "▁creating": 697,
+      "▁creativity": 698,
+      "▁data):": 699,
+      "▁data:": 700,
+      "▁data?": 701,
+      "▁datasets,": 702,
+      "▁days,": 703,
+      "▁de-link": 704,
+      "▁decay": 705,
+      "▁demonstrations": 706,
+      "▁demonstrations.": 707,
+      "▁dependencies": 708,
+      "▁deployment.": 709,
+      "▁described": 710,
+      "▁detection": 711,
+      "▁detection,": 712,
+      "▁developers": 713,
+      "▁developing": 714,
+      "▁development": 715,
+      "▁difference": 716,
+      "▁difficult": 717,
+      "▁directory": 718,
+      "▁distributed": 719,
+      "▁diverse": 720,
+      "▁diversity": 721,
+      "▁documents": 722,
+      "▁domain": 723,
+      "▁domain-specific": 724,
+      "▁dynamic": 725,
+      "▁easiest": 726,
+      "▁easy": 727,
+      "▁effectively.": 728,
+      "▁effects": 729,
+      "▁emergent": 730,
+      "▁encoding": 731,
+      "▁engineers.": 732,
+      "▁enough": 733,
+      "▁enterprises": 734,
+      "▁epochs": 735,
+      "▁evaluating": 736,
+      "▁excel": 737,
+      "▁expensive": 738,
+      "▁experience,": 739,
+      "▁experimentation": 740,
+      "▁explicitly": 741,
+      "▁fake": 742,
+      "▁few": 743,
+      "▁fine-tune": 744,
+      "▁first.": 745,
+      "▁flow": 746,
+      "▁found": 747,
+      "▁foundation": 748,
+      "▁frameworks": 749,
+      "▁fundamental": 750,
+      "▁gateway": 751,
+      "▁gathering": 752,
+      "▁gave": 753,
+      "▁generation": 754,
+      "▁generation,": 755,
+      "▁generation.": 756,
+      "▁generators,": 757,
+      "▁give": 758,
+      "▁goals": 759,
+      "▁going": 760,
+      "▁good": 761,
+      "▁ground": 762,
+      "▁groundbreaking": 763,
+      "▁had": 764,
+      "▁half": 765,
+      "▁handling": 766,
+      "▁hardware": 767,
+      "▁having": 768,
+      "▁helped": 769,
+      "▁heuristics": 770,
+      "▁high": 771,
+      "▁high-quality": 772,
+      "▁idea": 773,
+      "▁images,": 774,
+      "▁implementing": 775,
+      "▁important": 776,
+      "▁improved": 777,
+      "▁improvement": 778,
+      "▁included": 779,
+      "▁incredibly": 780,
+      "▁industry.": 781,
+      "▁inference": 782,
+      "▁inference.": 783,
+      "▁information.": 784,
+      "▁infrastructure,": 785,
+      "▁initial": 786,
+      "▁innovation.": 787,
+      "▁instructions": 788,
+      "▁interacts": 789,
+      "▁interesting": 790,
+      "▁intermediate": 791,
+      "▁internal": 792,
+      "▁introduction": 793,
+      "▁investment": 794,
+      "▁journey.": 795,
+      "▁just": 796,
+      "▁keep": 797,
+      "▁known": 798,
+      "▁laws.": 799,
+      "▁layer": 800,
+      "▁layers": 801,
+      "▁leads": 802,
+      "▁learns": 803,
+      "▁legal": 804,
+      "▁llm": 805,
+      "▁load": 806,
+      "▁look": 807,
+      "▁loss": 808,
+      "▁lot": 809,
+      "▁main": 810,
+      "▁making": 811,
+      "▁manner": 812,
+      "▁me,": 813,
+      "▁measure": 814,
+      "▁mechanism,": 815,
+      "▁memory": 816,
+      "▁metrics.": 817,
+      "▁millions": 818,
+      "▁mistakes,": 819,
+      "▁model:": 820,
+      "▁monitor": 821,
+      "▁must": 822,
+      "▁necessary": 823,
+      "▁needed": 824,
+      "▁neighbor": 825,
+      "▁network,": 826,
+      "▁now": 827,
+      "▁number:": 828,
+      "▁objective": 829,
+      "▁objective.": 830,
+      "▁once": 831,
+      "▁open": 832,
+      "▁optimal": 833,
+      "▁optimized": 834,
+      "▁option": 835,
+      "▁outcomes.": 836,
+      "▁output,": 837,
+      "▁padding": 838,
+      "▁parameters": 839,
+      "▁parameters,": 840,
+      "▁parameters.": 841,
+      "▁particularly": 842,
+      "▁passing": 843,
+      "▁path": 844,
+      "▁perform": 845,
+      "▁permitted": 846,
+      "▁personalized": 847,
+      "▁pip": 848,
+      "▁point": 849,
+      "▁practical": 850,
+      "▁pre-training": 851,
+      "▁predict": 852,
+      "▁preferences,": 853,
+      "▁prepared": 854,
+      "▁pretrained": 855,
+      "▁problems,": 856,
+      "▁produce": 857,
+      "▁products": 858,
+      "▁project": 859,
+      "▁prompt,": 860,
+      "▁proprietary": 861,
+      "▁provided": 862,
+      "▁provider,": 863,
+      "▁provides": 864,
+      "▁punctuation,": 865,
+      "▁pure": 866,
+      "▁quality": 867,
+      "▁questions,": 868,
+      "▁rate": 869,
+      "▁rate,": 870,
+      "▁raw": 871,
+      "▁real": 872,
+      "▁real-world": 873,
+      "▁related": 874,
+      "▁remote": 875,
+      "▁removing": 876,
+      "▁repeated": 877,
+      "▁requires": 878,
+      "▁residual": 879,
+      "▁result": 880,
+      "▁results,": 881,
+      "▁review": 882,
+      "▁right": 883,
+      "▁satisfies": 884,
+      "▁scope": 885,
+      "▁scratch": 886,
+      "▁second": 887,
+      "▁sent": 888,
+      "▁sentiment": 889,
+      "▁sequence.": 890,
+      "▁servers,": 891,
+      "▁sessions": 892,
+      "▁set.": 893,
+      "▁share": 894,
+      "▁should": 895,
+      "▁show": 896,
+      "▁showing": 897,
+      "▁significantly": 898,
+      "▁solutions": 899,
+      "▁solutions.": 900,
+      "▁something": 901,
+      "▁sophisticated": 902,
+      "▁speaks": 903,
+      "▁splits": 904,
+      "▁started": 905,
+      "▁starting": 906,
+      "▁step\"": 907,
+      "▁steps,": 908,
+      "▁storing": 909,
+      "▁strategies": 910,
+      "▁study": 911,
+      "▁styles": 912,
+      "▁suggests": 913,
+      "▁support.": 914,
+      "▁switch,": 915,
+      "▁system": 916,
+      "▁takes": 917,
+      "▁taking": 918,
+      "▁teams": 919,
+      "▁terms": 920,
+      "▁test": 921,
+      "▁things": 922,
+      "▁this.": 923,
+      "▁thorough": 924,
+      "▁those": 925,
+      "▁thumbs": 926,
+      "▁time.": 927,
+      "▁times": 928,
+      "▁to)": 929,
+      "▁to.": 930,
+      "▁today": 931,
+      "▁token": 932,
+      "▁tokens,": 933,
+      "▁tokens.": 934,
+      "▁toxicity": 935,
+      "▁trail": 936,
+      "▁train-your-own-model": 937,
+      "▁transformed": 938,
+      "▁translation": 939,
+      "▁trial-and-error,": 940,
+      "▁try": 941,
+      "▁tune": 942,
+      "▁tuning": 943,
+      "▁type": 944,
+      "▁types": 945,
+      "▁under": 946,
+      "▁understand,": 947,
+      "▁updated": 948,
+      "▁us": 949,
+      "▁use,": 950,
+      "▁useful.": 951,
+      "▁usually": 952,
+      "▁versatile": 953,
+      "▁via": 954,
+      "▁websites,": 955,
+      "▁weight": 956,
+      "▁went": 957,
+      "▁who": 958,
+      "▁words,": 959,
+      "▁world": 960,
+      "▁write": 961,
+      "▁writing": 962,
+      "▁yet": 963,
+      "▁“in": 964,
+      "▁#1": 965,
+      "▁#2": 966,
+      "▁#can": 967,
+      "▁$324,573": 968,
+      "▁$357,542": 969,
+      "▁$375,286": 970,
+      "▁$388,852": 971,
+      "▁$402,255": 972,
+      "▁(2022),": 973,
+      "▁(9,": 974,
+      "▁(AI)": 975,
+      "▁(AI),": 976,
+      "▁(AWS).": 977,
+      "▁(Auto-CoT)": 978,
+      "▁(BPE),": 979,
+      "▁(CoT)": 980,
+      "▁(GPU": 981,
+      "▁(Kojima": 982,
+      "▁(LLM)?": 983,
+      "▁(LLMs).": 984,
+      "▁(ML)": 985,
+      "▁(More": 986,
+      "▁(PII)": 987,
+      "▁(RLHF)": 988,
+      "▁(SGD)": 989,
+      "▁(Sales,": 990,
+      "▁(Sweeps)": 991,
+      "▁(decrease": 992,
+      "▁(from": 993,
+      "▁(graphics": 994,
+      "▁(grounded": 995,
+      "▁(grouping": 996,
+      "▁(in": 997,
+      "▁(including": 998,
+      "▁(making": 999,
+      "▁(not": 1000,
+      "▁(see": 1001,
+      "▁(splitting": 1002,
+      "▁(train": 1003,
+      "▁(using": 1004,
+      "▁(without": 1005,
+      "▁(writing": 1006,
+      "▁/": 1007,
+      "▁1": 1008,
+      "▁1)": 1009,
+      "▁1):": 1010,
+      "▁1-3": 1011,
+      "▁11": 1012,
+      "▁12,": 1013,
+      "▁13": 1014,
+      "▁13,": 1015,
+      "▁17B": 1016,
+      "▁2,": 1017,
+      "▁2021.": 1018,
+      "▁2022)": 1019,
+      "▁2024:": 1020,
+      "▁2026.6-layer.": 1021,
+      "▁25.": 1022,
+      "▁32,": 1023,
+      "▁397B": 1024,
+      "▁3:": 1025,
+      "▁4": 1026,
+      "▁4,": 1027,
+      "▁4:": 1028,
+      "▁5,": 1029,
+      "▁5:": 1030,
+      "▁6-layer": 1031,
+      "▁60": 1032,
+      "▁6:": 1033,
+      "▁7,": 1034,
+      "▁7.": 1035,
+      "▁7:": 1036,
+      "▁8,": 1037,
+      "▁8.": 1038,
+      "▁82,": 1039,
+      "▁9,": 1040,
+      "▁9.": 1041,
+      "▁:": 1042,
+      "▁ACME.": 1043,
+      "▁AI:": 1044,
+      "▁API,": 1045,
+      "▁APIs),": 1046,
+      "▁APIs?": 1047,
+      "▁APIs”": 1048,
+      "▁AWS": 1049,
+      "▁AWS.2": 1050,
+      "▁Account": 1051,
+      "▁Adjusting": 1052,
+      "▁Adventure.": 1053,
+      "▁Alibaba’s": 1054,
+      "▁Always": 1055,
+      "▁Amazon": 1056,
+      "▁Among": 1057,
+      "▁Analyze": 1058,
+      "▁And": 1059,
+      "▁Another": 1060,
+      "▁Anthropic": 1061,
+      "▁Anthropic,": 1062,
+      "▁Anthropic’s": 1063,
+      "▁Any": 1064,
+      "▁Apex": 1065,
+      "▁Applications": 1066,
+      "▁Architecture:": 1067,
+      "▁Artificial": 1068,
+      "▁As": 1069,
+      "▁Assemble": 1070,
+      "▁Audit": 1071,
+      "▁Auto-CoT": 1072,
+      "▁Auto-CoT,": 1073,
+      "▁AutoTokenizer": 1074,
+      "▁AutoTokenizer.from_pretrained('gpt-4')": 1075,
+      "▁Automatic": 1076,
+      "▁Azure": 1077,
+      "▁BERT,": 1078,
+      "▁BLEU": 1079,
+      "▁BLEU,": 1080,
+      "▁Balancing": 1081,
+      "▁Based": 1082,
+      "▁Basic": 1083,
+      "▁Batch": 1084,
+      "▁Batching": 1085,
+      "▁Be": 1086,
+      "▁Before": 1087,
+      "▁Better": 1088,
+      "▁Biases": 1089,
+      "▁Biases,": 1090,
+      "▁Bigger": 1091,
+      "▁Blackwell": 1092,
+      "▁BlueField-4": 1093,
+      "▁Bypass": 1094,
+      "▁Byte": 1095,
+      "▁CCPA,": 1096,
+      "▁CEO": 1097,
+      "▁CEO.": 1098,
+      "▁Campaign": 1099,
+      "▁Chain-of-Thought": 1100,
+      "▁Chat": 1101,
+      "▁Chrome.": 1102,
+      "▁Cleaning:": 1103,
+      "▁CoT": 1104,
+      "▁Code.": 1105,
+      "▁Cohere,": 1106,
+      "▁Colab": 1107,
+      "▁Collect": 1108,
+      "▁Commerce),": 1109,
+      "▁Community": 1110,
+      "▁Computational": 1111,
+      "▁Considerations:": 1112,
+      "▁Consistently": 1113,
+      "▁Contents": 1114,
+      "▁Convert": 1115,
+      "▁Cost": 1116,
+      "▁Crawl": 1117,
+      "▁Crawl,": 1118,
+      "▁Creating": 1119,
+      "▁Custom": 1120,
+      "▁DPU,": 1121,
+      "▁Data.gov": 1122,
+      "▁Dead:": 1123,
+      "▁Decide": 1124,
+      "▁Decrease": 1125,
+      "▁DeepSeek-R1": 1126,
+      "▁Define": 1127,
+      "▁Demasking": 1128,
+      "▁Demasking,": 1129,
+      "▁Depending": 1130,
+      "▁Deploy": 1131,
+      "▁Deployment": 1132,
+      "▁Detection": 1133,
+      "▁Detectors": 1134,
+      "▁Determine": 1135,
+      "▁Determining": 1136,
+      "▁DevOps": 1137,
+      "▁DevOps,": 1138,
+      "▁Developer": 1139,
+      "▁Developers": 1140,
+      "▁Different": 1141,
+      "▁Difficulty": 1142,
+      "▁Domain-specific": 1143,
+      "▁Don’t": 1144,
+      "▁Download": 1145,
+      "▁Drawing": 1146,
+      "▁Each": 1147,
+      "▁Edge,": 1148,
+      "▁Edition": 1149,
+      "▁Eliminate": 1150,
+      "▁Elite,": 1151,
+      "▁Engage": 1152,
+      "▁English": 1153,
+      "▁Ethernet": 1154,
+      "▁Ethical": 1155,
+      "▁Evaluate": 1156,
+      "▁Everyone": 1157,
+      "▁Experimenting": 1158,
+      "▁Expertise:": 1159,
+      "▁F1-score": 1160,
+      "▁FAQs": 1161,
+      "▁FP8": 1162,
+      "▁Face)": 1163,
+      "▁Face.": 1164,
+      "▁False.": 1165,
+      "▁Familiarity": 1166,
+      "▁FastAPI": 1167,
+      "▁Feed": 1168,
+      "▁Feed-forward": 1169,
+      "▁Feel": 1170,
+      "▁Final": 1171,
+      "▁Finally,": 1172,
+      "▁Fine": 1173,
+      "▁Fine-tune": 1174,
+      "▁Flask": 1175,
+      "▁Follow": 1176,
+      "▁Framework": 1177,
+      "▁Free,": 1178,
+      "▁Freeze": 1179,
+      "▁From": 1180,
+      "▁GDPR": 1181,
+      "▁GPT": 1182,
+      "▁GPT-3,": 1183,
+      "▁GPT-3.": 1184,
+      "▁GPT2LMHeadModel": 1185,
+      "▁GPT2LMHeadModel.from_pretrained('gpt-4')": 1186,
+      "▁GPU.": 1187,
+      "▁GPUs": 1188,
+      "▁GUI": 1189,
+      "▁Game": 1190,
+      "▁Getting": 1191,
+      "▁GitHub": 1192,
+      "▁Given": 1193,
+      "▁Google.": 1194,
+      "▁Google’s": 1195,
+      "▁Graph": 1196,
+      "▁Happy": 1197,
+      "▁Hard": 1198,
+      "▁Have": 1199,
+      "▁Hyperparameter": 1200,
+      "▁Hyperparameters": 1201,
+      "▁I'm": 1202,
+      "▁Image": 1203,
+      "▁Implementing": 1204,
+      "▁Improve": 1205,
+      "▁Infrastructure": 1206,
+      "▁Input": 1207,
+      "▁Instead,": 1208,
+      "▁Integrating": 1209,
+      "▁Intelligence": 1210,
+      "▁Introduced": 1211,
+      "▁Involvement:": 1212,
+      "▁IoT,": 1213,
+      "▁Is": 1214,
+      "▁It's": 1215,
+      "▁Kojima": 1216,
+      "▁LLM!": 1217,
+      "▁LLM:": 1218,
+      "▁LLMs,": 1219,
+      "▁LaMDA": 1220,
+      "▁Larger": 1221,
+      "▁Layer’s": 1222,
+      "▁Leading": 1223,
+      "▁Learning:": 1224,
+      "▁Libraries:": 1225,
+      "▁LinkedIn": 1226,
+      "▁Lisa": 1227,
+      "▁Llama": 1228,
+      "▁Llama,": 1229,
+      "▁Long": 1230,
+      "▁Loop:": 1231,
+      "▁Loss": 1232,
+      "▁Lower": 1233,
+      "▁Luckily,": 1234,
+      "▁Machine": 1235,
+      "▁Major": 1236,
+      "▁Make": 1237,
+      "▁Making": 1238,
+      "▁Marketing": 1239,
+      "▁Marketing,": 1240,
+      "▁Martinez,": 1241,
+      "▁Max": 1242,
+      "▁Medical": 1243,
+      "▁Mentor,": 1244,
+      "▁Meta": 1245,
+      "▁Metrics": 1246,
+      "▁Metrics:": 1247,
+      "▁Microsoft,": 1248,
+      "▁Mixture-of-Experts": 1249,
+      "▁MoE": 1250,
+      "▁Model.": 1251,
+      "▁Model:": 1252,
+      "▁Model?": 1253,
+      "▁Most": 1254,
+      "▁Multi-head": 1255,
+      "▁NLP.": 1256,
+      "▁NVIDIA": 1257,
+      "▁NVLink": 1258,
+      "▁Natural": 1259,
+      "▁New": 1260,
+      "▁Nobody": 1261,
+      "▁Normalization": 1262,
+      "▁Northern": 1263,
+      "▁Objective": 1264,
+      "▁Open-source": 1265,
+      "▁OpenAI": 1266,
+      "▁OpenML,": 1267,
+      "▁Our": 1268,
+      "▁Outfitters.": 1269,
+      "▁Output:": 1270,
+      "▁Own": 1271,
+      "▁PaLM": 1272,
+      "▁Pair": 1273,
+      "▁Peak,": 1274,
+      "▁Perplexity:": 1275,
+      "▁Pipeline": 1276,
+      "▁Platform": 1277,
+      "▁Plenty": 1278,
+      "▁Policy,": 1279,
+      "▁Popular": 1280,
+      "▁Pre-trained": 1281,
+      "▁Pre-training": 1282,
+      "▁Prepare": 1283,
+      "▁Preparing": 1284,
+      "▁Preprocess": 1285,
+      "▁Preprocessing": 1286,
+      "▁Preprocessing:": 1287,
+      "▁Pretrained": 1288,
+      "▁Pro,": 1289,
+      "▁Processing": 1290,
+      "▁Program)": 1291,
+      "▁Public": 1292,
+      "▁PyTorch": 1293,
+      "▁PyTorch,": 1294,
+      "▁PyTorch.": 1295,
+      "▁Python:": 1296,
+      "▁Question": 1297,
+      "▁Qwen3.5": 1298,
+      "▁Qwen3.5,": 1299,
+      "▁Qwen3.5-397B-A17B,": 1300,
+      "▁RAM,": 1301,
+      "▁RL": 1302,
+      "▁RLHF,": 1303,
+      "▁ROUGE:": 1304,
+      "▁Recently,": 1305,
+      "▁Recognize": 1306,
+      "▁Record": 1307,
+      "▁RefinedWeb": 1308,
+      "▁Reflecting": 1309,
+      "▁Regularization": 1310,
+      "▁Regularly": 1311,
+      "▁Reinforcement": 1312,
+      "▁Representative": 1313,
+      "▁Required": 1314,
+      "▁Required:": 1315,
+      "▁Resources:": 1316,
+      "▁Rubin-based": 1317,
+      "▁Rubin-era": 1318,
+      "▁SVP": 1319,
+      "▁Safeguard": 1320,
+      "▁Safeguards": 1321,
+      "▁Salesforce.": 1322,
+      "▁Savings": 1323,
+      "▁Schedule": 1324,
+      "▁Search,": 1325,
+      "▁Secure": 1326,
+      "▁Security": 1327,
+      "▁Select": 1328,
+      "▁Sentiment": 1329,
+      "▁Service,": 1330,
+      "▁Serving": 1331,
+      "▁Setting": 1332,
+      "▁Settings.": 1333,
+      "▁Setup:": 1334,
+      "▁Should": 1335,
+      "▁Simplify": 1336,
+      "▁Skills": 1337,
+      "▁Smaller": 1338,
+      "▁Smith,": 1339,
+      "▁So,": 1340,
+      "▁Source:": 1341,
+      "▁Sources:": 1342,
+      "▁Sourcing": 1343,
+      "▁Speaker,": 1344,
+      "▁Spectrum-6": 1345,
+      "▁Split": 1346,
+      "▁Starting": 1347,
+      "▁State": 1348,
+      "▁Stay": 1349,
+      "▁Steps": 1350,
+      "▁Stochastic": 1351,
+      "▁Strategic": 1352,
+      "▁SuperNIC,": 1353,
+      "▁Support:": 1354,
+      "▁T5": 1355,
+      "▁T5.": 1356,
+      "▁Table": 1357,
+      "▁Tailored": 1358,
+      "▁Task-specific": 1359,
+      "▁Technical": 1360,
+      "▁Techniques:": 1361,
+      "▁Tensor": 1362,
+      "▁Tenth": 1363,
+      "▁Tester": 1364,
+      "▁That’s": 1365,
+      "▁Then": 1366,
+      "▁Then,": 1367,
+      "▁Thoughts": 1368,
+      "▁Tier": 1369,
+      "▁Tokenization.": 1370,
+      "▁Tokenization:": 1371,
+      "▁Tokenize": 1372,
+      "▁Tokens": 1373,
+      "▁Toxicity": 1374,
+      "▁Track": 1375,
+      "▁Trail": 1376,
+      "▁Trainer(": 1377,
+      "▁TrainingArguments": 1378,
+      "▁TrainingArguments(": 1379,
+      "▁Transformers": 1380,
+      "▁Transformers,": 1381,
+      "▁Translation": 1382,
+      "▁Trusted": 1383,
+      "▁Truths": 1384,
+      "▁Tuning": 1385,
+      "▁Tuning:": 1386,
+      "▁Understanding": 1387,
+      "▁Undetectable": 1388,
+      "▁Up": 1389,
+      "▁Usage": 1390,
+      "▁Useful": 1391,
+      "▁Using": 1392,
+      "▁Validation:": 1393,
+      "▁Varying": 1394,
+      "▁Vera": 1395,
+      "▁Verify": 1396,
+      "▁Vision": 1397,
+      "▁Weave": 1398,
+      "▁Wei": 1399,
+      "▁Why": 1400,
+      "▁Why?": 1401,
+      "▁Word": 1402,
+      "▁WordPiece": 1403,
+      "▁Work": 1404,
+      "▁Wow!": 1405,
+      "▁abilities": 1406,
+      "▁abilities.": 1407,
+      "▁above": 1408,
+      "▁access.": 1409,
+      "▁account": 1410,
+      "▁accounts": 1411,
+      "▁achieving": 1412,
+      "▁acquired": 1413,
+      "▁act": 1414,
+      "▁activated": 1415,
+      "▁adapt": 1416,
+      "▁adaptation.": 1417,
+      "▁added": 1418,
+      "▁added,": 1419,
+      "▁additional": 1420,
+      "▁adjacent": 1421,
+      "▁adjusts": 1422,
+      "▁adopt": 1423,
+      "▁advancements": 1424,
+      "▁advantages": 1425,
+      "▁affect": 1426,
+      "▁agreements": 1427,
+      "▁aiming": 1428,
+      "▁algorithm": 1429,
+      "▁algorithms": 1430,
+      "▁aligning": 1431,
+      "▁although": 1432,
+      "▁amounts:": 1433,
+      "▁analysis": 1434,
+      "▁anymore.": 1435,
+      "▁app,": 1436,
+      "▁appears": 1437,
+      "▁apple,": 1438,
+      "▁apples,": 1439,
+      "▁application": 1440,
+      "▁applications,": 1441,
+      "▁applications?": 1442,
+      "▁approach:": 1443,
+      "▁approach?": 1444,
+      "▁approximately": 1445,
+      "▁architected": 1446,
+      "▁architectures,": 1447,
+      "▁are:": 1448,
+      "▁areas": 1449,
+      "▁args=training_args,": 1450,
+      "▁arguments,": 1451,
+      "▁arise": 1452,
+      "▁arises": 1453,
+      "▁article": 1454,
+      "▁article,": 1455,
+      "▁as:": 1456,
+      "▁ask": 1457,
+      "▁asking": 1458,
+      "▁assigns": 1459,
+      "▁assistance": 1460,
+      "▁assistant": 1461,
+      "▁associated": 1462,
+      "▁attempt": 1463,
+      "▁attend": 1464,
+      "▁attention.": 1465,
+      "▁audio,": 1466,
+      "▁auditing": 1467,
+      "▁authentication": 1468,
+      "▁authorized": 1469,
+      "▁authors": 1470,
+      "▁automatic": 1471,
+      "▁automating": 1472,
+      "▁availability": 1473,
+      "▁availability,": 1474,
+      "▁available": 1475,
+      "▁away": 1476,
+      "▁back-end": 1477,
+      "▁batching": 1478,
+      "▁be.": 1479,
+      "▁become": 1480,
+      "▁becomes": 1481,
+      "▁before.": 1482,
+      "▁belongs": 1483,
+      "▁below.": 1484,
+      "▁below:": 1485,
+      "▁benchmark": 1486,
+      "▁beneficial:": 1487,
+      "▁benefits": 1488,
+      "▁best": 1489,
+      "▁beyond": 1490,
+      "▁beyond.": 1491,
+      "▁bias": 1492,
+      "▁biggest": 1493,
+      "▁billions": 1494,
+      "▁bind": 1495,
+      "▁block": 1496,
+      "▁blocks": 1497,
+      "▁blog": 1498,
+      "▁books,": 1499,
+      "▁brand": 1500,
+      "▁break": 1501,
+      "▁breaking": 1502,
+      "▁builders": 1503,
+      "▁building": 1504,
+      "▁built": 1505,
+      "▁business": 1506,
+      "▁button": 1507,
+      "▁button,": 1508,
+      "▁buys": 1509,
+      "▁calls": 1510,
+      "▁calls.": 1511,
+      "▁came": 1512,
+      "▁can,": 1513,
+      "▁capabilities": 1514,
+      "▁capabilities.": 1515,
+      "▁captures": 1516,
+      "▁carefully": 1517,
+      "▁cases,": 1518,
+      "▁cases.": 1519,
+      "▁cater": 1520,
+      "▁certain": 1521,
+      "▁chain": 1522,
+      "▁chains.": 1523,
+      "▁challenge": 1524,
+      "▁changed": 1525,
+      "▁changing": 1526,
+      "▁characteristics": 1527,
+      "▁characters,": 1528,
+      "▁chart": 1529,
+      "▁chatbot": 1530,
+      "▁chatbot?": 1531,
+      "▁chatbots": 1532,
+      "▁chatbots,": 1533,
+      "▁chats.": 1534,
+      "▁checks": 1535,
+      "▁chip": 1536,
+      "▁chips.": 1537,
+      "▁choosing": 1538,
+      "▁chunks).": 1539,
+      "▁chunks.": 1540,
+      "▁claim": 1541,
+      "▁classification": 1542,
+      "▁claude@gmail.com": 1543,
+      "▁clean": 1544,
+      "▁cleaned": 1545,
+      "▁cleaner.": 1546,
+      "▁click.": 1547,
+      "▁clone": 1548,
+      "▁cloud,": 1549,
+      "▁cluster": 1550,
+      "▁clustering:": 1551,
+      "▁code,": 1552,
+      "▁code.": 1553,
+      "▁codesign": 1554,
+      "▁collaborate": 1555,
+      "▁combination": 1556,
+      "▁comments": 1557,
+      "▁commercial": 1558,
+      "▁committed": 1559,
+      "▁committing": 1560,
+      "▁common.": 1561,
+      "▁communication": 1562,
+      "▁communities": 1563,
+      "▁company,": 1564,
+      "▁company.": 1565,
+      "▁company?": 1566,
+      "▁compare": 1567,
+      "▁comparing": 1568,
+      "▁compass": 1569,
+      "▁compelling": 1570,
+      "▁complete": 1571,
+      "▁completely": 1572,
+      "▁complexity": 1573,
+      "▁compliance.": 1574,
+      "▁compliance:": 1575,
+      "▁complicated": 1576,
+      "▁complies": 1577,
+      "▁comply": 1578,
+      "▁components.": 1579,
+      "▁comprehend": 1580,
+      "▁comprehension.": 1581,
+      "▁comprehensive": 1582,
+      "▁computationally": 1583,
+      "▁compute,": 1584,
+      "▁computer,": 1585,
+      "▁conducting": 1586,
+      "▁configure": 1587,
+      "▁configured": 1588,
+      "▁configuring": 1589,
+      "▁connect": 1590,
+      "▁connection": 1591,
+      "▁connections": 1592,
+      "▁consistency,": 1593,
+      "▁consistency.": 1594,
+      "▁consistent": 1595,
+      "▁consistently": 1596,
+      "▁construct": 1597,
+      "▁consumer": 1598,
+      "▁consumption,": 1599,
+      "▁containerization": 1600,
+      "▁contains": 1601,
+      "▁content!": 1602,
+      "▁context": 1603,
+      "▁contexts": 1604,
+      "▁contextual": 1605,
+      "▁continues": 1606,
+      "▁continuing": 1607,
+      "▁continuously": 1608,
+      "▁convergence": 1609,
+      "▁converges": 1610,
+      "▁convert": 1611,
+      "▁converting": 1612,
+      "▁copy": 1613,
+      "▁copyright": 1614,
+      "▁cornerstone": 1615,
+      "▁costs.": 1616,
+      "▁countless": 1617,
+      "▁covered": 1618,
+      "▁crafted": 1619,
+      "▁created": 1620,
+      "▁creates": 1621,
+      "▁creation": 1622,
+      "▁creator": 1623,
+      "▁critical": 1624,
+      "▁criticism": 1625,
+      "▁crowded": 1626,
+      "▁crucial.": 1627,
+      "▁curated": 1628,
+      "▁curves": 1629,
+      "▁customization,": 1630,
+      "▁cut": 1631,
+      "▁cutting": 1632,
+      "▁data)": 1633,
+      "▁database": 1634,
+      "▁dataset,": 1635,
+      "▁datasets.": 1636,
+      "▁datasets.1": 1637,
+      "▁datasets:": 1638,
+      "▁daunting": 1639,
+      "▁deal": 1640,
+      "▁decide": 1641,
+      "▁decision": 1642,
+      "▁decision.": 1643,
+      "▁decisions": 1644,
+      "▁decrease": 1645,
+      "▁deeper": 1646,
+      "▁default": 1647,
+      "▁defined": 1648,
+      "▁defining": 1649,
+      "▁deliver": 1650,
+      "▁demands": 1651,
+      "▁demasking,": 1652,
+      "▁demonstration": 1653,
+      "▁demonstrations,": 1654,
+      "▁dense": 1655,
+      "▁dependencies.": 1656,
+      "▁deployed,": 1657,
+      "▁deployed.": 1658,
+      "▁deployment": 1659,
+      "▁derived": 1660,
+      "▁descent": 1661,
+      "▁describe": 1662,
+      "▁describes": 1663,
+      "▁design": 1664,
+      "▁desired": 1665,
+      "▁detect": 1666,
+      "▁detected": 1667,
+      "▁determines": 1668,
+      "▁development,": 1669,
+      "▁development.": 1670,
+      "▁diagnosis": 1671,
+      "▁did": 1672,
+      "▁differences": 1673,
+      "▁difficulties": 1674,
+      "▁direct": 1675,
+      "▁disadvantages.": 1676,
+      "▁discussed": 1677,
+      "▁discussions.": 1678,
+      "▁diverges": 1679,
+      "▁diverse,": 1680,
+      "▁diverse.": 1681,
+      "▁dividing": 1682,
+      "▁do,": 1683,
+      "▁documentation,": 1684,
+      "▁does.": 1685,
+      "▁doesn’t": 1686,
+      "▁doing": 1687,
+      "▁dollars.": 1688,
+      "▁domain-specific,": 1689,
+      "▁domain.": 1690,
+      "▁don't": 1691,
+      "▁downside": 1692,
+      "▁draft": 1693,
+      "▁drawbacks,": 1694,
+      "▁dropout": 1695,
+      "▁duplicates": 1696,
+      "▁duplicates,": 1697,
+      "▁during": 1698,
+      "▁dynamically": 1699,
+      "▁earlier": 1700,
+      "▁earlier.": 1701,
+      "▁early": 1702,
+      "▁easily": 1703,
+      "▁economical": 1704,
+      "▁economics": 1705,
+      "▁efficiency,": 1706,
+      "▁efficiency:": 1707,
+      "▁efficient.": 1708,
+      "▁efficiently.": 1709,
+      "▁effort": 1710,
+      "▁efforts": 1711,
+      "▁elements": 1712,
+      "▁elevate": 1713,
+      "▁eliminate": 1714,
+      "▁else": 1715,
+      "▁embedding": 1716,
+      "▁embedding,": 1717,
+      "▁embeddings": 1718,
+      "▁embeddings,": 1719,
+      "▁emerged": 1720,
+      "▁emergence": 1721,
+      "▁emphasize": 1722,
+      "▁enable": 1723,
+      "▁enabled": 1724,
+      "▁encoding.": 1725,
+      "▁encourages": 1726,
+      "▁end": 1727,
+      "▁end,": 1728,
+      "▁endeavor.": 1729,
+      "▁enforce": 1730,
+      "▁engineers": 1731,
+      "▁engineers,": 1732,
+      "▁enough:": 1733,
+      "▁enriched": 1734,
+      "▁ensured": 1735,
+      "▁ensures": 1736,
+      "▁entail": 1737,
+      "▁entails": 1738,
+      "▁enter": 1739,
+      "▁enterprise": 1740,
+      "▁enterprises,": 1741,
+      "▁entirely,": 1742,
+      "▁environment": 1743,
+      "▁environment,": 1744,
+      "▁environment.": 1745,
+      "▁epochs.": 1746,
+      "▁equipped": 1747,
+      "▁errors.": 1748,
+      "▁essence": 1749,
+      "▁essentially": 1750,
+      "▁establish": 1751,
+      "▁etc.": 1752,
+      "▁etc.)": 1753,
+      "▁etc...": 1754,
+      "▁ethical": 1755,
+      "▁ethically": 1756,
+      "▁eval_dataset=eval_dataset": 1757,
+      "▁evaluate": 1758,
+      "▁evaluated": 1759,
+      "▁evaluations": 1760,
+      "▁ever": 1761,
+      "▁every": 1762,
+      "▁everyone": 1763,
+      "▁example.": 1764,
+      "▁examples,": 1765,
+      "▁examples.": 1766,
+      "▁excited": 1767,
+      "▁execution,": 1768,
+      "▁exhibit": 1769,
+      "▁expected": 1770,
+      "▁expensive,": 1771,
+      "▁experience": 1772,
+      "▁expertise": 1773,
+      "▁expertise,": 1774,
+      "▁experts": 1775,
+      "▁exploring,": 1776,
+      "▁exposing": 1777,
+      "▁extend": 1778,
+      "▁extensive": 1779,
+      "▁extensive,": 1780,
+      "▁extent.": 1781,
+      "▁extraction.": 1782,
+      "▁eye": 1783,
+      "▁fact,": 1784,
+      "▁fails,": 1785,
+      "▁fails.": 1786,
+      "▁fair": 1787,
+      "▁faithfulness,": 1788,
+      "▁fall": 1789,
+      "▁familiar": 1790,
+      "▁familiarity": 1791,
+      "▁far": 1792,
+      "▁fascinates": 1793,
+      "▁fascinating": 1794,
+      "▁fast.": 1795,
+      "▁fastest": 1796,
+      "▁fed": 1797,
+      "▁feed": 1798,
+      "▁feed-forward": 1799,
+      "▁feedback.": 1800,
+      "▁feeding": 1801,
+      "▁feel": 1802,
+      "▁few-shot": 1803,
+      "▁fewer": 1804,
+      "▁field": 1805,
+      "▁field.": 1806,
+      "▁fields": 1807,
+      "▁finance,": 1808,
+      "▁financial": 1809,
+      "▁fine-tuned": 1810,
+      "▁fine-tuning,": 1811,
+      "▁fine-tuning.": 1812,
+      "▁finished": 1813,
+      "▁fit.": 1814,
+      "▁fix": 1815,
+      "▁fixing": 1816,
+      "▁flagged": 1817,
+      "▁flows,": 1818,
+      "▁fluency,": 1819,
+      "▁focus": 1820,
+      "▁follow": 1821,
+      "▁follow-up": 1822,
+      "▁follow.": 1823,
+      "▁followed": 1824,
+      "▁form": 1825,
+      "▁format": 1826,
+      "▁formats": 1827,
+      "▁forms,": 1828,
+      "▁forums": 1829,
+      "▁foundational": 1830,
+      "▁four": 1831,
+      "▁frameworks,": 1832,
+      "▁free": 1833,
+      "▁freedom": 1834,
+      "▁frequent": 1835,
+      "▁friendly": 1836,
+      "▁from.": 1837,
+      "▁full-scale": 1838,
+      "▁function.": 1839,
+      "▁functions": 1840,
+      "▁fundamentals": 1841,
+      "▁future.": 1842,
+      "▁game-changer": 1843,
+      "▁game-changing": 1844,
+      "▁game.": 1845,
+      "▁gateway:": 1846,
+      "▁gather": 1847,
+      "▁gauge": 1848,
+      "▁gemini@gmail.com": 1849,
+      "▁general": 1850,
+      "▁general-purpose": 1851,
+      "▁generalized": 1852,
+      "▁generates": 1853,
+      "▁generating": 1854,
+      "▁generator": 1855,
+      "▁generator,": 1856,
+      "▁generators": 1857,
+      "▁generic,": 1858,
+      "▁git": 1859,
+      "▁given": 1860,
+      "▁go": 1861,
+      "▁goal,": 1862,
+      "▁goal.": 1863,
+      "▁golden": 1864,
+      "▁good?": 1865,
+      "▁governed": 1866,
+      "▁gpt-4@gmail.com": 1867,
+      "▁gradient": 1868,
+      "▁granting": 1869,
+      "▁graphical": 1870,
+      "▁great": 1871,
+      "▁great,": 1872,
+      "▁greatest": 1873,
+      "▁grow": 1874,
+      "▁guides": 1875,
+      "▁hand,": 1876,
+      "▁hand-crafting": 1877,
+      "▁happy": 1878,
+      "▁hardware.": 1879,
+      "▁hardware–software": 1880,
+      "▁harmful": 1881,
+      "▁harmonize": 1882,
+      "▁haven’t": 1883,
+      "▁heads": 1884,
+      "▁healthcare,": 1885,
+      "▁heart,": 1886,
+      "▁heavily": 1887,
+      "▁here\",": 1888,
+      "▁here.": 1889,
+      "▁heterogeneous": 1890,
+      "▁high-volume,": 1891,
+      "▁highly": 1892,
+      "▁hot": 1893,
+      "▁hours": 1894,
+      "▁hours,": 1895,
+      "▁how.": 1896,
+      "▁https://github.com/SreeEswaran/Train-your-LLM": 1897,
+      "▁human-labeled": 1898,
+      "▁human.": 1899,
+      "▁humans": 1900,
+      "▁hundred": 1901,
+      "▁hybrid": 1902,
+      "▁hyperparameters,": 1903,
+      "▁hyperparameters.": 1904,
+      "▁i.e.,": 1905,
+      "▁identifiable": 1906,
+      "▁identify": 1907,
+      "▁identity": 1908,
+      "▁if:": 1909,
+      "▁illustrate": 1910,
+      "▁illustrates": 1911,
+      "▁immediately": 1912,
+      "▁immense": 1913,
+      "▁implement": 1914,
+      "▁implemented": 1915,
+      "▁implications": 1916,
+      "▁important.": 1917,
+      "▁imported": 1918,
+      "▁impressive": 1919,
+      "▁improvement.": 1920,
+      "▁improvements,": 1921,
+      "▁in-context": 1922,
+      "▁in.": 1923,
+      "▁include:": 1924,
+      "▁incognito": 1925,
+      "▁incorporate": 1926,
+      "▁incorrect!": 1927,
+      "▁increase": 1928,
+      "▁increased": 1929,
+      "▁increasing": 1930,
+      "▁increasingly": 1931,
+      "▁indicates": 1932,
+      "▁individual": 1933,
+      "▁industries": 1934,
+      "▁industries.": 1935,
+      "▁industry": 1936,
+      "▁industry,": 1937,
+      "▁industry:": 1938,
+      "▁influenced": 1939,
+      "▁information,": 1940,
+      "▁informed": 1941,
+      "▁infrastructure.": 1942,
+      "▁innovation": 1943,
+      "▁innovation,": 1944,
+      "▁innovative": 1945,
+      "▁input.": 1946,
+      "▁inputs": 1947,
+      "▁inside": 1948,
+      "▁insightful,": 1949,
+      "▁insights": 1950,
+      "▁insights!": 1951,
+      "▁installed.": 1952,
+      "▁instance,": 1953,
+      "▁instances.": 1954,
+      "▁instantiated": 1955,
+      "▁integrate": 1956,
+      "▁integrating": 1957,
+      "▁intensive": 1958,
+      "▁interaction,": 1959,
+      "▁intimidating": 1960,
+      "▁intricacy": 1961,
+      "▁intricate": 1962,
+      "▁introduced": 1963,
+      "▁introduces": 1964,
+      "▁invest": 1965,
+      "▁involve": 1966,
+      "▁involved": 1967,
+      "▁involving": 1968,
+      "▁irrelevant": 1969,
+      "▁is.": 1970,
+      "▁isn’t": 1971,
+      "▁issue:": 1972,
+      "▁it.": 1973,
+      "▁iteratively,": 1974,
+      "▁job.": 1975,
+      "▁joining": 1976,
+      "▁journey,": 1977,
+      "▁kept": 1978,
+      "▁key.": 1979,
+      "▁kind": 1980,
+      "▁kinds": 1981,
+      "▁knowledge": 1982,
+      "▁knowledge.": 1983,
+      "▁knowledgeable": 1984,
+      "▁labor": 1985,
+      "▁labs,": 1986,
+      "▁language,": 1987,
+      "▁language-related": 1988,
+      "▁larger": 1989,
+      "▁last": 1990,
+      "▁late": 1991,
+      "▁later": 1992,
+      "▁later.)": 1993,
+      "▁laws": 1994,
+      "▁layer,": 1995,
+      "▁leading": 1996,
+      "▁leaked": 1997,
+      "▁learning.": 1998,
+      "▁least": 1999,
+      "▁left.": 2000,
+      "▁lemmatization": 2001,
+      "▁length": 2002,
+      "▁length),": 2003,
+      "▁length.": 2004,
+      "▁lengthy": 2005,
+      "▁less": 2006,
+      "▁let": 2007,
+      "▁level,": 2008,
+      "▁level.": 2009,
+      "▁leverage": 2010,
+      "▁leverages": 2011,
+      "▁leveraging": 2012,
+      "▁libraries.": 2013,
+      "▁library": 2014,
+      "▁licensing": 2015,
+      "▁lies": 2016,
+      "▁light": 2017,
+      "▁like:": 2018,
+      "▁liked": 2019,
+      "▁likely": 2020,
+      "▁limitations": 2021,
+      "▁limited": 2022,
+      "▁line": 2023,
+      "▁line,": 2024,
+      "▁linear": 2025,
+      "▁lines:": 2026,
+      "▁lingo": 2027,
+      "▁list": 2028,
+      "▁literature,": 2029,
+      "▁llama@gmail.com": 2030,
+      "▁logging": 2031,
+      "▁logging.": 2032,
+      "▁logging_dir='./logs',": 2033,
+      "▁logs": 2034,
+      "▁logs,": 2035,
+      "▁long-context": 2036,
+      "▁loop": 2037,
+      "▁lower-cost": 2038,
+      "▁lowercase": 2039,
+      "▁lowercase,": 2040,
+      "▁magnitude": 2041,
+      "▁maintaining": 2042,
+      "▁maintenance": 2043,
+      "▁makes": 2044,
+      "▁manage": 2045,
+      "▁manageable": 2046,
+      "▁manner.": 2047,
+      "▁manually": 2048,
+      "▁map": 2049,
+      "▁market": 2050,
+      "▁market.": 2051,
+      "▁masking": 2052,
+      "▁masking,": 2053,
+      "▁mass": 2054,
+      "▁master": 2055,
+      "▁match": 2056,
+      "▁math": 2057,
+      "▁mathematical": 2058,
+      "▁mathematicians.": 2059,
+      "▁mathematics": 2060,
+      "▁mathematics.": 2061,
+      "▁matter": 2062,
+      "▁matter.": 2063,
+      "▁maximize": 2064,
+      "▁media.": 2065,
+      "▁mentioned": 2066,
+      "▁met": 2067,
+      "▁meticulous": 2068,
+      "▁metric": 2069,
+      "▁millions.": 2070,
+      "▁min": 2071,
+      "▁mind": 2072,
+      "▁mind.": 2073,
+      "▁mindful": 2074,
+      "▁mini-batches": 2075,
+      "▁miniature": 2076,
+      "▁minimize": 2077,
+      "▁missing": 2078,
+      "▁mission)": 2079,
+      "▁mistakes": 2080,
+      "▁mistral": 2081,
+      "▁mistral@gmail.com": 2082,
+      "▁misuse": 2083,
+      "▁mitigate": 2084,
+      "▁mitigation": 2085,
+      "▁mix": 2086,
+      "▁mixture-of-experts": 2087,
+      "▁mixture-of-experts,": 2088,
+      "▁model=model,": 2089,
+      "▁model?": 2090,
+      "▁modeling": 2091,
+      "▁models)": 2092,
+      "▁models:": 2093,
+      "▁modern": 2094,
+      "▁modify": 2095,
+      "▁modules": 2096,
+      "▁money": 2097,
+      "▁months.": 2098,
+      "▁moral.": 2099,
+      "▁more!": 2100,
+      "▁more,": 2101,
+      "▁much": 2102,
+      "▁multi-head": 2103,
+      "▁multiple": 2104,
+      "▁native": 2105,
+      "▁naturally": 2106,
+      "▁nature": 2107,
+      "▁nature.": 2108,
+      "▁needed.": 2109,
+      "▁network": 2110,
+      "▁network.": 2111,
+      "▁networking.": 2112,
+      "▁news": 2113,
+      "▁next.": 2114,
+      "▁non-linear": 2115,
+      "▁non-negotiable.": 2116,
+      "▁normalized,": 2117,
+      "▁note": 2118,
+      "▁notice": 2119,
+      "▁now,": 2120,
+      "▁nuances": 2121,
+      "▁nudging": 2122,
+      "▁num_train_epochs=3,": 2123,
+      "▁numerical": 2124,
+      "▁numerous": 2125,
+      "▁observability": 2126,
+      "▁offensive": 2127,
+      "▁offering": 2128,
+      "▁offers": 2129,
+      "▁on.": 2130,
+      "▁one.": 2131,
+      "▁ones": 2132,
+      "▁ongoing": 2133,
+      "▁open-weight": 2134,
+      "▁openai,": 2135,
+      "▁opportunities": 2136,
+      "▁opposing": 2137,
+      "▁optimization": 2138,
+      "▁optimize": 2139,
+      "▁opting": 2140,
+      "▁option:": 2141,
+      "▁options.": 2142,
+      "▁orders:": 2143,
+      "▁organization’s": 2144,
+      "▁original": 2145,
+      "▁other,": 2146,
+      "▁others)": 2147,
+      "▁otherwise": 2148,
+      "▁output.": 2149,
+      "▁output_dir='./results',": 2150,
+      "▁outputs.": 2151,
+      "▁overall": 2152,
+      "▁overfitting": 2153,
+      "▁page,": 2154,
+      "▁paper": 2155,
+      "▁papers.": 2156,
+      "▁paragraphs": 2157,
+      "▁parallel,": 2158,
+      "▁particular": 2159,
+      "▁partition": 2160,
+      "▁partner": 2161,
+      "▁parts,": 2162,
+      "▁parts:": 2163,
+      "▁passed": 2164,
+      "▁patience": 2165,
+      "▁pattern": 2166,
+      "▁patterns": 2167,
+      "▁patterns.": 2168,
+      "▁peculiarities": 2169,
+      "▁people": 2170,
+      "▁per": 2171,
+      "▁per_device_eval_batch_size=4,": 2172,
+      "▁per_device_train_batch_size=4,": 2173,
+      "▁perfect": 2174,
+      "▁perform,": 2175,
+      "▁performance.3": 2176,
+      "▁performs:": 2177,
+      "▁perplexity": 2178,
+      "▁perplexity,": 2179,
+      "▁persist": 2180,
+      "▁personal": 2181,
+      "▁pertinent": 2182,
+      "▁phase,": 2183,
+      "▁philosophical": 2184,
+      "▁phone": 2185,
+      "▁pieces": 2186,
+      "▁pipelines,": 2187,
+      "▁place": 2188,
+      "▁placeholder": 2189,
+      "▁plan": 2190,
+      "▁plans": 2191,
+      "▁platform,": 2192,
+      "▁platform.": 2193,
+      "▁platforms": 2194,
+      "▁play.": 2195,
+      "▁please": 2196,
+      "▁plethora": 2197,
+      "▁plug": 2198,
+      "▁poetry": 2199,
+      "▁policies": 2200,
+      "▁possibilities": 2201,
+      "▁possible.": 2202,
+      "▁post,": 2203,
+      "▁postdoc": 2204,
+      "▁posts": 2205,
+      "▁potent": 2206,
+      "▁potentially": 2207,
+      "▁power,": 2208,
+      "▁power.": 2209,
+      "▁powerful,": 2210,
+      "▁powering": 2211,
+      "▁powers": 2212,
+      "▁practice,": 2213,
+      "▁practices": 2214,
+      "▁pre-processed": 2215,
+      "▁prebuilt": 2216,
+      "▁precision": 2217,
+      "▁precision,": 2218,
+      "▁predicting": 2219,
+      "▁predictions,": 2220,
+      "▁preference": 2221,
+      "▁preferences.": 2222,
+      "▁prepared,": 2223,
+      "▁preprocess": 2224,
+      "▁preprocessed": 2225,
+      "▁presented": 2226,
+      "▁prevent": 2227,
+      "▁price,": 2228,
+      "▁primarily": 2229,
+      "▁principle": 2230,
+      "▁principles": 2231,
+      "▁privacy.": 2232,
+      "▁privilege:": 2233,
+      "▁probably": 2234,
+      "▁problem": 2235,
+      "▁problems.": 2236,
+      "▁procedures.": 2237,
+      "▁processes.5": 2238,
+      "▁processor.": 2239,
+      "▁producing": 2240,
+      "▁product": 2241,
+      "▁production-level": 2242,
+      "▁productivity.": 2243,
+      "▁programmed": 2244,
+      "▁programming": 2245,
+      "▁progressing": 2246,
+      "▁project,": 2247,
+      "▁projects,": 2248,
+      "▁projects.": 2249,
+      "▁promotes": 2250,
+      "▁propose": 2251,
+      "▁proposes": 2252,
+      "▁protect": 2253,
+      "▁proven": 2254,
+      "▁provider": 2255,
+      "▁providers,": 2256,
+      "▁providing": 2257,
+      "▁purpose": 2258,
+      "▁purposes.": 2259,
+      "▁quality.": 2260,
+      "▁quantity.": 2261,
+      "▁rack-scale": 2262,
+      "▁raise": 2263,
+      "▁raises": 2264,
+      "▁rates:": 2265,
+      "▁rationale": 2266,
+      "▁reaches": 2267,
+      "▁read": 2268,
+      "▁ready.": 2269,
+      "▁real-time": 2270,
+      "▁realistic": 2271,
+      "▁reality": 2272,
+      "▁really": 2273,
+      "▁reason": 2274,
+      "▁reasoning.4": 2275,
+      "▁reasons": 2276,
+      "▁recall,": 2277,
+      "▁received": 2278,
+      "▁recent": 2279,
+      "▁recently": 2280,
+      "▁recognition.": 2281,
+      "▁recognize": 2282,
+      "▁reconsider": 2283,
+      "▁records": 2284,
+      "▁record’": 2285,
+      "▁reduce": 2286,
+      "▁reducing": 2287,
+      "▁refine": 2288,
+      "▁reflects": 2289,
+      "▁regard": 2290,
+      "▁regarding": 2291,
+      "▁regardless": 2292,
+      "▁regulations": 2293,
+      "▁regulations,": 2294,
+      "▁relations,": 2295,
+      "▁relationship": 2296,
+      "▁released": 2297,
+      "▁reliable": 2298,
+      "▁rely": 2299,
+      "▁remove": 2300,
+      "▁repairman,": 2301,
+      "▁repairman.": 2302,
+      "▁repeatedly": 2303,
+      "▁replaced": 2304,
+      "▁replaces": 2305,
+      "▁replicas,": 2306,
+      "▁report": 2307,
+      "▁report:": 2308,
+      "▁reportedly": 2309,
+      "▁repository:": 2310,
+      "▁representation,": 2311,
+      "▁representation.": 2312,
+      "▁representations.": 2313,
+      "▁representative": 2314,
+      "▁request": 2315,
+      "▁request.": 2316,
+      "▁required.": 2317,
+      "▁research-level": 2318,
+      "▁reshaped": 2319,
+      "▁resource": 2320,
+      "▁resources.": 2321,
+      "▁responding.": 2322,
+      "▁response": 2323,
+      "▁responsibilities.": 2324,
+      "▁responsible": 2325,
+      "▁restores": 2326,
+      "▁restrict": 2327,
+      "▁restricted": 2328,
+      "▁resulting": 2329,
+      "▁results.": 2330,
+      "▁retention": 2331,
+      "▁retention:": 2332,
+      "▁retraining": 2333,
+      "▁return_tensors='pt')": 2334,
+      "▁reward": 2335,
+      "▁rewarding": 2336,
+      "▁robust": 2337,
+      "▁route": 2338,
+      "▁rules": 2339,
+      "▁run": 2340,
+      "▁run.": 2341,
+      "▁runs.": 2342,
+      "▁samples": 2343,
+      "▁sampling:": 2344,
+      "▁satisfactory": 2345,
+      "▁satisfactory,": 2346,
+      "▁saving": 2347,
+      "▁scalable": 2348,
+      "▁scale": 2349,
+      "▁scaled": 2350,
+      "▁scheduler": 2351,
+      "▁scholarly": 2352,
+      "▁scientific": 2353,
+      "▁score": 2354,
+      "▁score,": 2355,
+      "▁scratch,": 2356,
+      "▁scrub": 2357,
+      "▁search": 2358,
+      "▁search,": 2359,
+      "▁secure": 2360,
+      "▁secured": 2361,
+      "▁security:": 2362,
+      "▁seemingly": 2363,
+      "▁seems": 2364,
+      "▁select": 2365,
+      "▁selected": 2366,
+      "▁selected.": 2367,
+      "▁selection": 2368,
+      "▁separate": 2369,
+      "▁series": 2370,
+      "▁server.": 2371,
+      "▁servers": 2372,
+      "▁service": 2373,
+      "▁service,": 2374,
+      "▁services,": 2375,
+      "▁session": 2376,
+      "▁sets": 2377,
+      "▁setup,": 2378,
+      "▁shaped": 2379,
+      "▁shared": 2380,
+      "▁sharing": 2381,
+      "▁shed": 2382,
+      "▁short": 2383,
+      "▁shown": 2384,
+      "▁shows": 2385,
+      "▁signals": 2386,
+      "▁simplify": 2387,
+      "▁simply": 2388,
+      "▁since": 2389,
+      "▁sits": 2390,
+      "▁six": 2391,
+      "▁size)": 2392,
+      "▁sizes": 2393,
+      "▁sizes:": 2394,
+      "▁skill.": 2395,
+      "▁skills": 2396,
+      "▁skills.": 2397,
+      "▁smart": 2398,
+      "▁snippets": 2399,
+      "▁social": 2400,
+      "▁software": 2401,
+      "▁solid": 2402,
+      "▁solutions,": 2403,
+      "▁sophistication,": 2404,
+      "▁sourced": 2405,
+      "▁space.": 2406,
+      "▁sparse": 2407,
+      "▁speak": 2408,
+      "▁specialize": 2409,
+      "▁specifications": 2410,
+      "▁specificity.": 2411,
+      "▁specified": 2412,
+      "▁speed.": 2413,
+      "▁speedups.": 2414,
+      "▁spend": 2415,
+      "▁spikes)": 2416,
+      "▁spread": 2417,
+      "▁stabilize": 2418,
+      "▁stages:": 2419,
+      "▁stand": 2420,
+      "▁standardize": 2421,
+      "▁standards": 2422,
+      "▁stands": 2423,
+      "▁start.": 2424,
+      "▁start?": 2425,
+      "▁static": 2426,
+      "▁statistical": 2427,
+      "▁stay": 2428,
+      "▁stemming.": 2429,
+      "▁steps).": 2430,
+      "▁steps.": 2431,
+      "▁stop": 2432,
+      "▁stopped": 2433,
+      "▁storage,": 2434,
+      "▁store": 2435,
+      "▁stored": 2436,
+      "▁stories": 2437,
+      "▁stories,": 2438,
+      "▁stories.": 2439,
+      "▁strategy,": 2440,
+      "▁strength": 2441,
+      "▁structure": 2442,
+      "▁styles,": 2443,
+      "▁suboptimal": 2444,
+      "▁subsequent": 2445,
+      "▁subsets": 2446,
+      "▁substantial": 2447,
+      "▁substantially": 2448,
+      "▁subtleties": 2449,
+      "▁subword": 2450,
+      "▁subwords.": 2451,
+      "▁success": 2452,
+      "▁sufficient": 2453,
+      "▁sufficiently": 2454,
+      "▁suitable": 2455,
+      "▁summarization": 2456,
+      "▁summarization.": 2457,
+      "▁summarizing": 2458,
+      "▁supercomputer": 2459,
+      "▁supercomputers.": 2460,
+      "▁supervised": 2461,
+      "▁supports": 2462,
+      "▁sure": 2463,
+      "▁surpass": 2464,
+      "▁symbols.": 2465,
+      "▁system.": 2466,
+      "▁systematic": 2467,
+      "▁systems,": 2468,
+      "▁tailoring": 2469,
+      "▁taken": 2470,
+      "▁talent,": 2471,
+      "▁talked": 2472,
+      "▁target": 2473,
+      "▁task,": 2474,
+      "▁task-specific": 2475,
+      "▁teaching": 2476,
+      "▁team,": 2477,
+      "▁technical": 2478,
+      "▁techniques)": 2479,
+      "▁techniques.": 2480,
+      "▁technologies": 2481,
+      "▁templates": 2482,
+      "▁tension": 2483,
+      "▁terminology,": 2484,
+      "▁tests.": 2485,
+      "▁text-generation": 2486,
+      "▁texting": 2487,
+      "▁texts": 2488,
+      "▁texts,": 2489,
+      "▁that’s": 2490,
+      "▁them.": 2491,
+      "▁there": 2492,
+      "▁thing": 2493,
+      "▁third-party": 2494,
+      "▁thoughts": 2495,
+      "▁thousands": 2496,
+      "▁through.": 2497,
+      "▁tight": 2498,
+      "▁time,": 2499,
+      "▁to:": 2500,
+      "▁today's": 2501,
+      "▁today,": 2502,
+      "▁token?”": 2503,
+      "▁tokenization.": 2504,
+      "▁tokenize": 2505,
+      "▁tokenizer": 2506,
+      "▁tokenizer(\"Your": 2507,
+      "▁tokenizing": 2508,
+      "▁tokens)": 2509,
+      "▁tone": 2510,
+      "▁tone,": 2511,
+      "▁took": 2512,
+      "▁tools,": 2513,
+      "▁top": 2514,
+      "▁topic": 2515,
+      "▁topics,": 2516,
+      "▁topics.": 2517,
+      "▁torch": 2518,
+      "▁touch": 2519,
+      "▁tracing": 2520,
+      "▁traditional": 2521,
+      "▁trail:": 2522,
+      "▁train),": 2523,
+      "▁train_dataset=train_dataset,": 2524,
+      "▁trained.": 2525,
+      "▁trainer": 2526,
+      "▁trainer.train()": 2527,
+      "▁training?": 2528,
+      "▁training_args": 2529,
+      "▁transformation": 2530,
+      "▁translate": 2531,
+      "▁translating": 2532,
+      "▁translation.": 2533,
+      "▁transmits": 2534,
+      "▁trillion": 2535,
+      "▁truly": 2536,
+      "▁trusted": 2537,
+      "▁trying": 2538,
+      "▁tweak": 2539,
+      "▁types.": 2540,
+      "▁typically": 2541,
+      "▁typos,": 2542,
+      "▁undergraduate": 2543,
+      "▁understand.": 2544,
+      "▁understanding.": 2545,
+      "▁undertaking.": 2546,
+      "▁unit)": 2547,
+      "▁units": 2548,
+      "▁units),": 2549,
+      "▁unlocks": 2550,
+      "▁unnecessary": 2551,
+      "▁unpublished,": 2552,
+      "▁unseen": 2553,
+      "▁unsolved": 2554,
+      "▁until": 2555,
+      "▁up,": 2556,
+      "▁up/down": 2557,
+      "▁updates": 2558,
+      "▁updating": 2559,
+      "▁upfront,": 2560,
+      "▁usage": 2561,
+      "▁used.": 2562,
+      "▁users’": 2563,
+      "▁uses": 2564,
+      "▁utilization,": 2565,
+      "▁utilized": 2566,
+      "▁valuable": 2567,
+      "▁valuable,": 2568,
+      "▁values.": 2569,
+      "▁variables": 2570,
+      "▁variations.": 2571,
+      "▁vast": 2572,
+      "▁vector": 2573,
+      "▁versatile.": 2574,
+      "▁versatility": 2575,
+      "▁version": 2576,
+      "▁very": 2577,
+      "▁video,": 2578,
+      "▁visualize": 2579,
+      "▁vitality": 2580,
+      "▁volume": 2581,
+      "▁wanted": 2582,
+      "▁wants": 2583,
+      "▁warmup": 2584,
+      "▁warmup_steps=500,": 2585,
+      "▁wasn’t": 2586,
+      "▁way": 2587,
+      "▁way,": 2588,
+      "▁way.": 2589,
+      "▁ways": 2590,
+      "▁website,": 2591,
+      "▁weeks": 2592,
+      "▁weeks.": 2593,
+      "▁weight_decay=0.01,": 2594,
+      "▁well-formatted.": 2595,
+      "▁well-known": 2596,
+      "▁well:": 2597,
+      "▁while": 2598,
+      "▁whole": 2599,
+      "▁wide": 2600,
+      "▁widespread.": 2601,
+      "▁with,": 2602,
+      "▁with?": 2603,
+      "▁within": 2604,
+      "▁word,": 2605,
+      "▁word.": 2606,
+      "▁words.": 2607,
+      "▁work,": 2608,
+      "▁workflow.": 2609,
+      "▁workflows": 2610,
+      "▁working": 2611,
+      "▁works": 2612,
+      "▁works:": 2613,
+      "▁worry,": 2614,
+      "▁worthwhile": 2615,
+      "▁wouldn’t": 2616,
+      "▁writers,": 2617,
+      "▁wrong": 2618,
+      "▁year.": 2619,
+      "▁years.": 2620,
+      "▁you,": 2621,
+      "▁zero": 2622,
+      "▁zero-shot": 2623,
+      "▁–": 2624,
+      "▁“large”": 2625,
+      "▁“use": 2626,
+      "▁“what’s": 2627
+    },
+    "merges": []
+  }
+}

tokenizer_config.json

@@ -0,0 +1,20 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "extra_special_tokens": [
+    "<pad>",
+    "</s>",
+    "<s>",
+    "<maskSub>",
+    "Question:",
+    "Réponse:"
+  ],
+  "is_local": true,
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<pad>",
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "vocab_file": "tokenizer-sora/tokenizer.model"
+}