upload models at 2500 iterations

config.json

@@ -0,0 +1,32 @@
+{
+  "_name_or_path": "facebook/galactica-1.3b",
+  "_remove_final_layer_norm": false,
+  "activation_dropout": 0.0,
+  "activation_function": "gelu",
+  "architectures": [
+    "OPTForCausalLM"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 0,
+  "do_layer_norm_before": true,
+  "dropout": 0.1,
+  "enable_bias": true,
+  "eos_token_id": 2,
+  "ffn_dim": 8192,
+  "hidden_size": 2048,
+  "init_std": 0.02,
+  "layer_norm_elementwise_affine": true,
+  "layerdrop": 0.0,
+  "learned_embeddings": true,
+  "max_position_embeddings": 2048,
+  "model_type": "opt",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 24,
+  "pad_token_id": 1,
+  "scale_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.25.1",
+  "use_cache": true,
+  "vocab_size": 50002,
+  "word_embed_proj_dim": 2048
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37f3def097356b3be26f8a392ccd4c39aa7ea999f705035efdf31dc8a728c486
+size 2630533369

special_tokens_map.json

@@ -0,0 +1,8 @@
+{
+  "additional_special_tokens": [
+    "<question>",
+    "<answer>"
+  ],
+  "eos_token": "</s>",
+  "pad_token": "<pad>"
+}

tokenizer_config.json

@@ -0,0 +1,6 @@
+{
+  "model_max_length": 1000000000000000019884624838656,
+  "name_or_path": "facebook/galactica-1.3b",
+  "special_tokens_map_file": "/content/tokenizer/special_tokens_map.json",
+  "tokenizer_class": "PreTrainedTokenizerFast"
+}

trainer_state.json

@@ -0,0 +1,1696 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.6367949393779653,
+  "global_step": 2500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.599516044570672e-07,
+      "loss": 2.3879,
+      "step": 10
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.6830783438602115e-07,
+      "loss": 2.3525,
+      "step": 20
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 5.316921656139787e-07,
+      "loss": 2.2352,
+      "step": 30
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 5.766640643149751e-07,
+      "loss": 2.2506,
+      "step": 40
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 6.115469789851804e-07,
+      "loss": 2.182,
+      "step": 50
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 6.400483955429327e-07,
+      "loss": 2.1697,
+      "step": 60
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 6.641459998837218e-07,
+      "loss": 2.1285,
+      "step": 70
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 6.850202942439291e-07,
+      "loss": 2.1158,
+      "step": 80
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.034327267708902e-07,
+      "loss": 2.1371,
+      "step": 90
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.199032089141344e-07,
+      "loss": 2.1271,
+      "step": 100
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.348025723496237e-07,
+      "loss": 2.101,
+      "step": 110
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.484046254718867e-07,
+      "loss": 2.1176,
+      "step": 120
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.60917301394197e-07,
+      "loss": 2.0376,
+      "step": 130
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.725022298126757e-07,
+      "loss": 2.1025,
+      "step": 140
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.832875401420919e-07,
+      "loss": 2.1164,
+      "step": 150
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.933765241728831e-07,
+      "loss": 2.0634,
+      "step": 160
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.028536679096446e-07,
+      "loss": 2.0979,
+      "step": 170
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.117889566998443e-07,
+      "loss": 2.0704,
+      "step": 180
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.202410148252902e-07,
+      "loss": 2.0474,
+      "step": 190
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 8.282594388430883e-07,
+      "loss": 2.0705,
+      "step": 200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 8.358865610406332e-07,
+      "loss": 2.0443,
+      "step": 210
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 8.431588022785777e-07,
+      "loss": 2.1009,
+      "step": 220
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 8.501077238654499e-07,
+      "loss": 2.0686,
+      "step": 230
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 8.567608554008407e-07,
+      "loss": 2.0261,
+      "step": 240
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 8.631423535132934e-07,
+      "loss": 2.0785,
+      "step": 250
+    },
+    {
+      "epoch": 0.26,
+      "eval_webgpt_accuracy": 0.5190179277190983,
+      "eval_webgpt_loss": 2.201171875,
+      "eval_webgpt_runtime": 429.5981,
+      "eval_webgpt_samples_per_second": 9.115,
+      "eval_webgpt_steps_per_second": 0.761,
+      "step": 250
+    },
+    {
+      "epoch": 0.26,
+      "eval_prompt_dialogue_accuracy": 0.5585842097604062,
+      "eval_prompt_dialogue_loss": 1.857421875,
+      "eval_prompt_dialogue_runtime": 1052.8834,
+      "eval_prompt_dialogue_samples_per_second": 9.791,
+      "eval_prompt_dialogue_steps_per_second": 0.817,
+      "step": 250
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 8.692735313231511e-07,
+      "loss": 2.0753,
+      "step": 260
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 8.751732879278018e-07,
+      "loss": 2.0926,
+      "step": 270
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 8.808584597416296e-07,
+      "loss": 2.0557,
+      "step": 280
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 8.863441101555992e-07,
+      "loss": 1.969,
+      "step": 290
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 8.916437700710459e-07,
+      "loss": 2.049,
+      "step": 300
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 8.967696389785231e-07,
+      "loss": 2.063,
+      "step": 310
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 9.017327541018372e-07,
+      "loss": 2.0055,
+      "step": 320
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 9.065431335065352e-07,
+      "loss": 2.0109,
+      "step": 330
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 9.112098978385987e-07,
+      "loss": 2.0601,
+      "step": 340
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 9.157413744118348e-07,
+      "loss": 2.0343,
+      "step": 350
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 9.201451866287983e-07,
+      "loss": 2.0123,
+      "step": 360
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 9.244283311473209e-07,
+      "loss": 2.0567,
+      "step": 370
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 9.285972447542441e-07,
+      "loss": 2.0171,
+      "step": 380
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 9.326578625511087e-07,
+      "loss": 2.0287,
+      "step": 390
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 9.366156687720423e-07,
+      "loss": 2.0357,
+      "step": 400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 9.404757413257927e-07,
+      "loss": 2.0556,
+      "step": 410
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 9.442427909695872e-07,
+      "loss": 2.0166,
+      "step": 420
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 9.479211958729468e-07,
+      "loss": 2.028,
+      "step": 430
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 9.515150322075316e-07,
+      "loss": 2.0044,
+      "step": 440
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 9.550281012990034e-07,
+      "loss": 2.0301,
+      "step": 450
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 9.584639537944038e-07,
+      "loss": 2.0321,
+      "step": 460
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 9.61825911230254e-07,
+      "loss": 1.9889,
+      "step": 470
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 9.651170853297947e-07,
+      "loss": 2.0272,
+      "step": 480
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 9.683403953103762e-07,
+      "loss": 2.0319,
+      "step": 490
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 9.714985834422475e-07,
+      "loss": 2.0251,
+      "step": 500
+    },
+    {
+      "epoch": 0.53,
+      "eval_webgpt_accuracy": 0.522718880616189,
+      "eval_webgpt_loss": 2.181640625,
+      "eval_webgpt_runtime": 433.6644,
+      "eval_webgpt_samples_per_second": 9.03,
+      "eval_webgpt_steps_per_second": 0.754,
+      "step": 500
+    },
+    {
+      "epoch": 0.53,
+      "eval_prompt_dialogue_accuracy": 0.5654158849986294,
+      "eval_prompt_dialogue_loss": 1.810546875,
+      "eval_prompt_dialogue_runtime": 1051.7495,
+      "eval_prompt_dialogue_samples_per_second": 9.802,
+      "eval_prompt_dialogue_steps_per_second": 0.818,
+      "step": 500
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 9.745942290665561e-07,
+      "loss": 2.0089,
+      "step": 510
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 9.776297612521052e-07,
+      "loss": 2.0385,
+      "step": 520
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 9.80607470246476e-07,
+      "loss": 1.9865,
+      "step": 530
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 9.835295178567558e-07,
+      "loss": 1.9859,
+      "step": 540
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 9.863979468777368e-07,
+      "loss": 2.0115,
+      "step": 550
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 9.892146896705837e-07,
+      "loss": 2.004,
+      "step": 560
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 9.919815759822018e-07,
+      "loss": 2.0179,
+      "step": 570
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 9.94700340084553e-07,
+      "loss": 2.0083,
+      "step": 580
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 9.97372627303682e-07,
+      "loss": 2.0227,
+      "step": 590
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 9.999999999999997e-07,
+      "loss": 1.9993,
+      "step": 600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 9.959893048128342e-07,
+      "loss": 2.0316,
+      "step": 610
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 9.915329768270946e-07,
+      "loss": 1.9877,
+      "step": 620
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 9.870766488413547e-07,
+      "loss": 1.9862,
+      "step": 630
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 9.82620320855615e-07,
+      "loss": 2.0064,
+      "step": 640
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 9.781639928698751e-07,
+      "loss": 2.0837,
+      "step": 650
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 9.737076648841355e-07,
+      "loss": 1.9865,
+      "step": 660
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 9.692513368983956e-07,
+      "loss": 2.0137,
+      "step": 670
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 9.64795008912656e-07,
+      "loss": 2.0423,
+      "step": 680
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 9.603386809269163e-07,
+      "loss": 1.992,
+      "step": 690
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 9.558823529411764e-07,
+      "loss": 1.9865,
+      "step": 700
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.514260249554367e-07,
+      "loss": 2.0152,
+      "step": 710
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.46969696969697e-07,
+      "loss": 2.0183,
+      "step": 720
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.425133689839572e-07,
+      "loss": 2.0077,
+      "step": 730
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 9.380570409982175e-07,
+      "loss": 1.9948,
+      "step": 740
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 9.336007130124776e-07,
+      "loss": 2.0004,
+      "step": 750
+    },
+    {
+      "epoch": 0.79,
+      "eval_webgpt_accuracy": 0.5246072651558654,
+      "eval_webgpt_loss": 2.171875,
+      "eval_webgpt_runtime": 427.6682,
+      "eval_webgpt_samples_per_second": 9.157,
+      "eval_webgpt_steps_per_second": 0.765,
+      "step": 750
+    },
+    {
+      "epoch": 0.79,
+      "eval_prompt_dialogue_accuracy": 0.5691780443314355,
+      "eval_prompt_dialogue_loss": 1.78515625,
+      "eval_prompt_dialogue_runtime": 1055.9887,
+      "eval_prompt_dialogue_samples_per_second": 9.762,
+      "eval_prompt_dialogue_steps_per_second": 0.814,
+      "step": 750
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.29144385026738e-07,
+      "loss": 2.0092,
+      "step": 760
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.246880570409982e-07,
+      "loss": 1.9648,
+      "step": 770
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.202317290552585e-07,
+      "loss": 2.0095,
+      "step": 780
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 9.157754010695186e-07,
+      "loss": 1.9805,
+      "step": 790
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 9.11319073083779e-07,
+      "loss": 1.999,
+      "step": 800
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 9.068627450980392e-07,
+      "loss": 2.0067,
+      "step": 810
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 9.024064171122994e-07,
+      "loss": 1.9908,
+      "step": 820
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 8.979500891265597e-07,
+      "loss": 1.9952,
+      "step": 830
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 8.9349376114082e-07,
+      "loss": 1.9924,
+      "step": 840
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 8.890374331550802e-07,
+      "loss": 2.022,
+      "step": 850
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 8.845811051693403e-07,
+      "loss": 2.0003,
+      "step": 860
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 8.801247771836007e-07,
+      "loss": 2.0315,
+      "step": 870
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 8.756684491978609e-07,
+      "loss": 1.9881,
+      "step": 880
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 8.712121212121211e-07,
+      "loss": 1.946,
+      "step": 890
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 8.667557932263814e-07,
+      "loss": 2.0094,
+      "step": 900
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 8.622994652406417e-07,
+      "loss": 1.9853,
+      "step": 910
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 8.578431372549019e-07,
+      "loss": 1.9821,
+      "step": 920
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 8.533868092691621e-07,
+      "loss": 1.9795,
+      "step": 930
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 8.489304812834224e-07,
+      "loss": 1.9971,
+      "step": 940
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 8.444741532976827e-07,
+      "loss": 2.0708,
+      "step": 950
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 8.400178253119428e-07,
+      "loss": 1.9623,
+      "step": 960
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 8.355614973262032e-07,
+      "loss": 1.9546,
+      "step": 970
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 8.311051693404634e-07,
+      "loss": 1.9878,
+      "step": 980
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 8.266488413547237e-07,
+      "loss": 1.9843,
+      "step": 990
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 8.221925133689838e-07,
+      "loss": 1.9807,
+      "step": 1000
+    },
+    {
+      "epoch": 1.05,
+      "eval_webgpt_accuracy": 0.5253906990392405,
+      "eval_webgpt_loss": 2.16796875,
+      "eval_webgpt_runtime": 428.1397,
+      "eval_webgpt_samples_per_second": 9.147,
+      "eval_webgpt_steps_per_second": 0.764,
+      "step": 1000
+    },
+    {
+      "epoch": 1.05,
+      "eval_prompt_dialogue_accuracy": 0.5718087356550049,
+      "eval_prompt_dialogue_loss": 1.7685546875,
+      "eval_prompt_dialogue_runtime": 1050.2103,
+      "eval_prompt_dialogue_samples_per_second": 9.816,
+      "eval_prompt_dialogue_steps_per_second": 0.819,
+      "step": 1000
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 8.177361853832442e-07,
+      "loss": 1.9773,
+      "step": 1010
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 8.132798573975044e-07,
+      "loss": 2.0443,
+      "step": 1020
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 8.088235294117646e-07,
+      "loss": 1.9528,
+      "step": 1030
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 8.043672014260249e-07,
+      "loss": 1.9365,
+      "step": 1040
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 7.999108734402852e-07,
+      "loss": 1.9579,
+      "step": 1050
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 7.954545454545454e-07,
+      "loss": 1.974,
+      "step": 1060
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 7.909982174688057e-07,
+      "loss": 1.9537,
+      "step": 1070
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 7.865418894830659e-07,
+      "loss": 1.9373,
+      "step": 1080
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 7.820855614973262e-07,
+      "loss": 1.9555,
+      "step": 1090
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 7.776292335115863e-07,
+      "loss": 1.973,
+      "step": 1100
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 7.731729055258467e-07,
+      "loss": 1.9401,
+      "step": 1110
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 7.687165775401069e-07,
+      "loss": 1.9508,
+      "step": 1120
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 7.642602495543672e-07,
+      "loss": 1.934,
+      "step": 1130
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 7.598039215686273e-07,
+      "loss": 1.9325,
+      "step": 1140
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 7.553475935828877e-07,
+      "loss": 1.9914,
+      "step": 1150
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 7.508912655971479e-07,
+      "loss": 1.978,
+      "step": 1160
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 7.464349376114081e-07,
+      "loss": 1.9442,
+      "step": 1170
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 7.419786096256684e-07,
+      "loss": 1.9598,
+      "step": 1180
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 7.375222816399287e-07,
+      "loss": 1.9529,
+      "step": 1190
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 7.330659536541889e-07,
+      "loss": 1.9675,
+      "step": 1200
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 7.286096256684492e-07,
+      "loss": 1.9433,
+      "step": 1210
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 7.241532976827094e-07,
+      "loss": 1.9002,
+      "step": 1220
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 7.196969696969697e-07,
+      "loss": 1.9349,
+      "step": 1230
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 7.152406417112298e-07,
+      "loss": 1.9414,
+      "step": 1240
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 7.107843137254902e-07,
+      "loss": 1.9555,
+      "step": 1250
+    },
+    {
+      "epoch": 1.32,
+      "eval_webgpt_accuracy": 0.526057600642472,
+      "eval_webgpt_loss": 2.1640625,
+      "eval_webgpt_runtime": 428.0509,
+      "eval_webgpt_samples_per_second": 9.148,
+      "eval_webgpt_steps_per_second": 0.764,
+      "step": 1250
+    },
+    {
+      "epoch": 1.32,
+      "eval_prompt_dialogue_accuracy": 0.5739159961148795,
+      "eval_prompt_dialogue_loss": 1.755859375,
+      "eval_prompt_dialogue_runtime": 1048.4647,
+      "eval_prompt_dialogue_samples_per_second": 9.832,
+      "eval_prompt_dialogue_steps_per_second": 0.82,
+      "step": 1250
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 7.063279857397504e-07,
+      "loss": 1.983,
+      "step": 1260
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 7.018716577540107e-07,
+      "loss": 1.983,
+      "step": 1270
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 6.974153297682709e-07,
+      "loss": 1.9449,
+      "step": 1280
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 6.929590017825312e-07,
+      "loss": 1.9886,
+      "step": 1290
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 6.885026737967914e-07,
+      "loss": 1.9334,
+      "step": 1300
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 6.840463458110517e-07,
+      "loss": 1.9732,
+      "step": 1310
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 6.795900178253119e-07,
+      "loss": 1.9457,
+      "step": 1320
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 6.751336898395722e-07,
+      "loss": 1.9691,
+      "step": 1330
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 6.706773618538324e-07,
+      "loss": 2.0199,
+      "step": 1340
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 6.662210338680927e-07,
+      "loss": 1.9492,
+      "step": 1350
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 6.617647058823529e-07,
+      "loss": 1.9553,
+      "step": 1360
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 6.573083778966132e-07,
+      "loss": 1.9766,
+      "step": 1370
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 6.528520499108734e-07,
+      "loss": 1.9496,
+      "step": 1380
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 6.483957219251337e-07,
+      "loss": 1.9677,
+      "step": 1390
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 6.439393939393939e-07,
+      "loss": 1.9043,
+      "step": 1400
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 6.394830659536542e-07,
+      "loss": 1.9571,
+      "step": 1410
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 6.350267379679144e-07,
+      "loss": 1.9631,
+      "step": 1420
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 6.305704099821747e-07,
+      "loss": 1.9587,
+      "step": 1430
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 6.261140819964349e-07,
+      "loss": 1.9621,
+      "step": 1440
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 6.216577540106952e-07,
+      "loss": 1.932,
+      "step": 1450
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 6.172014260249554e-07,
+      "loss": 1.9682,
+      "step": 1460
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 6.127450980392157e-07,
+      "loss": 1.9636,
+      "step": 1470
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 6.082887700534759e-07,
+      "loss": 1.9538,
+      "step": 1480
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 6.038324420677362e-07,
+      "loss": 1.9676,
+      "step": 1490
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 5.998217468805704e-07,
+      "loss": 1.9459,
+      "step": 1500
+    },
+    {
+      "epoch": 1.58,
+      "eval_webgpt_accuracy": 0.526490033682041,
+      "eval_webgpt_loss": 2.162109375,
+      "eval_webgpt_runtime": 435.1199,
+      "eval_webgpt_samples_per_second": 9.0,
+      "eval_webgpt_steps_per_second": 0.752,
+      "step": 1500
+    },
+    {
+      "epoch": 1.58,
+      "eval_prompt_dialogue_accuracy": 0.5751027402282565,
+      "eval_prompt_dialogue_loss": 1.748046875,
+      "eval_prompt_dialogue_runtime": 1051.1439,
+      "eval_prompt_dialogue_samples_per_second": 9.807,
+      "eval_prompt_dialogue_steps_per_second": 0.818,
+      "step": 1500
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 5.953654188948306e-07,
+      "loss": 1.9592,
+      "step": 1510
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 5.909090909090909e-07,
+      "loss": 1.9291,
+      "step": 1520
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 5.864527629233511e-07,
+      "loss": 1.9082,
+      "step": 1530
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 5.819964349376115e-07,
+      "loss": 1.9357,
+      "step": 1540
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 5.775401069518716e-07,
+      "loss": 1.9502,
+      "step": 1550
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 5.730837789661319e-07,
+      "loss": 1.9892,
+      "step": 1560
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 5.686274509803921e-07,
+      "loss": 1.983,
+      "step": 1570
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 5.641711229946524e-07,
+      "loss": 1.9745,
+      "step": 1580
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 5.597147950089126e-07,
+      "loss": 1.9541,
+      "step": 1590
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 5.552584670231729e-07,
+      "loss": 1.9482,
+      "step": 1600
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 5.508021390374332e-07,
+      "loss": 1.9341,
+      "step": 1610
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 5.463458110516934e-07,
+      "loss": 1.971,
+      "step": 1620
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 5.418894830659536e-07,
+      "loss": 1.9179,
+      "step": 1630
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 5.374331550802139e-07,
+      "loss": 1.9049,
+      "step": 1640
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 5.329768270944741e-07,
+      "loss": 1.9407,
+      "step": 1650
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 5.285204991087344e-07,
+      "loss": 1.9338,
+      "step": 1660
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 5.240641711229946e-07,
+      "loss": 1.9874,
+      "step": 1670
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 5.19607843137255e-07,
+      "loss": 1.9699,
+      "step": 1680
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 5.151515151515151e-07,
+      "loss": 1.9128,
+      "step": 1690
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 5.106951871657754e-07,
+      "loss": 1.9382,
+      "step": 1700
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 5.062388591800356e-07,
+      "loss": 1.9514,
+      "step": 1710
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 5.01782531194296e-07,
+      "loss": 1.9499,
+      "step": 1720
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 4.973262032085561e-07,
+      "loss": 1.9458,
+      "step": 1730
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 4.928698752228163e-07,
+      "loss": 1.9095,
+      "step": 1740
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 4.884135472370767e-07,
+      "loss": 1.9237,
+      "step": 1750
+    },
+    {
+      "epoch": 1.85,
+      "eval_webgpt_accuracy": 0.5268101464515921,
+      "eval_webgpt_loss": 2.16015625,
+      "eval_webgpt_runtime": 427.8942,
+      "eval_webgpt_samples_per_second": 9.152,
+      "eval_webgpt_steps_per_second": 0.764,
+      "step": 1750
+    },
+    {
+      "epoch": 1.85,
+      "eval_prompt_dialogue_accuracy": 0.5762533856613786,
+      "eval_prompt_dialogue_loss": 1.740234375,
+      "eval_prompt_dialogue_runtime": 1049.6135,
+      "eval_prompt_dialogue_samples_per_second": 9.822,
+      "eval_prompt_dialogue_steps_per_second": 0.819,
+      "step": 1750
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 4.839572192513369e-07,
+      "loss": 1.9181,
+      "step": 1760
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 4.795008912655971e-07,
+      "loss": 1.9393,
+      "step": 1770
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 4.750445632798574e-07,
+      "loss": 1.9278,
+      "step": 1780
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 4.705882352941176e-07,
+      "loss": 1.972,
+      "step": 1790
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 4.6613190730837784e-07,
+      "loss": 1.9545,
+      "step": 1800
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 4.616755793226381e-07,
+      "loss": 1.9324,
+      "step": 1810
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 4.5721925133689835e-07,
+      "loss": 1.9856,
+      "step": 1820
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 4.5276292335115863e-07,
+      "loss": 1.9013,
+      "step": 1830
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 4.4830659536541886e-07,
+      "loss": 1.9001,
+      "step": 1840
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 4.4385026737967914e-07,
+      "loss": 1.9326,
+      "step": 1850
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 4.3939393939393937e-07,
+      "loss": 1.9358,
+      "step": 1860
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 4.3493761140819965e-07,
+      "loss": 2.0019,
+      "step": 1870
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 4.304812834224599e-07,
+      "loss": 1.941,
+      "step": 1880
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 4.260249554367201e-07,
+      "loss": 1.9325,
+      "step": 1890
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 4.2112299465240644e-07,
+      "loss": 2.0644,
+      "step": 1900
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 4.1666666666666667e-07,
+      "loss": 1.9158,
+      "step": 1910
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 4.122103386809269e-07,
+      "loss": 1.9409,
+      "step": 1920
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 4.0775401069518717e-07,
+      "loss": 1.926,
+      "step": 1930
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 4.032976827094474e-07,
+      "loss": 1.9212,
+      "step": 1940
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 3.988413547237077e-07,
+      "loss": 1.9745,
+      "step": 1950
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 3.943850267379679e-07,
+      "loss": 1.9079,
+      "step": 1960
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 3.899286987522282e-07,
+      "loss": 1.9336,
+      "step": 1970
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 3.854723707664884e-07,
+      "loss": 1.9242,
+      "step": 1980
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 3.8101604278074864e-07,
+      "loss": 1.9357,
+      "step": 1990
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 3.765597147950089e-07,
+      "loss": 1.9036,
+      "step": 2000
+    },
+    {
+      "epoch": 2.11,
+      "eval_webgpt_accuracy": 0.5268452465359726,
+      "eval_webgpt_loss": 2.158203125,
+      "eval_webgpt_runtime": 428.2029,
+      "eval_webgpt_samples_per_second": 9.145,
+      "eval_webgpt_steps_per_second": 0.764,
+      "step": 2000
+    },
+    {
+      "epoch": 2.11,
+      "eval_prompt_dialogue_accuracy": 0.5769595660938633,
+      "eval_prompt_dialogue_loss": 1.7373046875,
+      "eval_prompt_dialogue_runtime": 1052.6454,
+      "eval_prompt_dialogue_samples_per_second": 9.793,
+      "eval_prompt_dialogue_steps_per_second": 0.817,
+      "step": 2000
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 3.7210338680926915e-07,
+      "loss": 1.8764,
+      "step": 2010
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 3.6764705882352943e-07,
+      "loss": 1.9615,
+      "step": 2020
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 3.6319073083778966e-07,
+      "loss": 1.914,
+      "step": 2030
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 3.5873440285204994e-07,
+      "loss": 1.8962,
+      "step": 2040
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 3.542780748663101e-07,
+      "loss": 1.8999,
+      "step": 2050
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 3.4982174688057034e-07,
+      "loss": 1.9456,
+      "step": 2060
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 3.453654188948306e-07,
+      "loss": 1.9347,
+      "step": 2070
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 3.4090909090909085e-07,
+      "loss": 1.9102,
+      "step": 2080
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 3.3645276292335113e-07,
+      "loss": 1.9355,
+      "step": 2090
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 3.3199643493761136e-07,
+      "loss": 1.9204,
+      "step": 2100
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 3.2754010695187164e-07,
+      "loss": 1.9396,
+      "step": 2110
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 3.2308377896613187e-07,
+      "loss": 1.9057,
+      "step": 2120
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 3.1862745098039215e-07,
+      "loss": 1.9241,
+      "step": 2130
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 3.141711229946524e-07,
+      "loss": 1.9383,
+      "step": 2140
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 3.097147950089126e-07,
+      "loss": 1.9197,
+      "step": 2150
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 3.052584670231729e-07,
+      "loss": 1.8983,
+      "step": 2160
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 3.008021390374331e-07,
+      "loss": 1.9677,
+      "step": 2170
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 2.963458110516934e-07,
+      "loss": 1.9104,
+      "step": 2180
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 2.918894830659536e-07,
+      "loss": 1.8994,
+      "step": 2190
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 2.874331550802139e-07,
+      "loss": 1.9157,
+      "step": 2200
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 2.829768270944741e-07,
+      "loss": 1.918,
+      "step": 2210
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 2.7852049910873435e-07,
+      "loss": 1.9053,
+      "step": 2220
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 2.7406417112299463e-07,
+      "loss": 1.8939,
+      "step": 2230
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 2.6960784313725486e-07,
+      "loss": 1.8959,
+      "step": 2240
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 2.6515151515151514e-07,
+      "loss": 1.9326,
+      "step": 2250
+    },
+    {
+      "epoch": 2.37,
+      "eval_webgpt_accuracy": 0.5270740990861342,
+      "eval_webgpt_loss": 2.158203125,
+      "eval_webgpt_runtime": 429.6721,
+      "eval_webgpt_samples_per_second": 9.114,
+      "eval_webgpt_steps_per_second": 0.761,
+      "step": 2250
+    },
+    {
+      "epoch": 2.37,
+      "eval_prompt_dialogue_accuracy": 0.577335217985265,
+      "eval_prompt_dialogue_loss": 1.7333984375,
+      "eval_prompt_dialogue_runtime": 1050.2434,
+      "eval_prompt_dialogue_samples_per_second": 9.816,
+      "eval_prompt_dialogue_steps_per_second": 0.819,
+      "step": 2250
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 2.6069518716577537e-07,
+      "loss": 1.8933,
+      "step": 2260
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 2.5623885918003565e-07,
+      "loss": 1.9186,
+      "step": 2270
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 2.517825311942959e-07,
+      "loss": 1.9149,
+      "step": 2280
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 2.473262032085561e-07,
+      "loss": 1.9288,
+      "step": 2290
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 2.428698752228164e-07,
+      "loss": 1.9082,
+      "step": 2300
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 2.3841354723707664e-07,
+      "loss": 1.9195,
+      "step": 2310
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 2.339572192513369e-07,
+      "loss": 1.9262,
+      "step": 2320
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 2.2950089126559712e-07,
+      "loss": 1.9096,
+      "step": 2330
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 2.2504456327985737e-07,
+      "loss": 1.913,
+      "step": 2340
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 2.2058823529411763e-07,
+      "loss": 1.9242,
+      "step": 2350
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 2.1613190730837788e-07,
+      "loss": 1.9175,
+      "step": 2360
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 2.1167557932263814e-07,
+      "loss": 1.8917,
+      "step": 2370
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 2.072192513368984e-07,
+      "loss": 1.9024,
+      "step": 2380
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 2.0276292335115864e-07,
+      "loss": 1.9243,
+      "step": 2390
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1.9830659536541887e-07,
+      "loss": 1.9269,
+      "step": 2400
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1.9385026737967912e-07,
+      "loss": 1.961,
+      "step": 2410
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.8939393939393938e-07,
+      "loss": 1.9414,
+      "step": 2420
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 1.8493761140819963e-07,
+      "loss": 1.9306,
+      "step": 2430
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 1.804812834224599e-07,
+      "loss": 1.9385,
+      "step": 2440
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.7602495543672014e-07,
+      "loss": 1.9304,
+      "step": 2450
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 1.715686274509804e-07,
+      "loss": 1.9214,
+      "step": 2460
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 1.6711229946524065e-07,
+      "loss": 1.9854,
+      "step": 2470
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 1.6265597147950088e-07,
+      "loss": 1.91,
+      "step": 2480
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 1.5819964349376113e-07,
+      "loss": 1.9317,
+      "step": 2490
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 1.5463458110516933e-07,
+      "loss": 1.9087,
+      "step": 2500
+    },
+    {
+      "epoch": 2.64,
+      "eval_webgpt_accuracy": 0.5270825231063856,
+      "eval_webgpt_loss": 2.15625,
+      "eval_webgpt_runtime": 426.8235,
+      "eval_webgpt_samples_per_second": 9.175,
+      "eval_webgpt_steps_per_second": 0.766,
+      "step": 2500
+    },
+    {
+      "epoch": 2.64,
+      "eval_prompt_dialogue_accuracy": 0.5776905643690232,
+      "eval_prompt_dialogue_loss": 1.7314453125,
+      "eval_prompt_dialogue_runtime": 1054.4289,
+      "eval_prompt_dialogue_samples_per_second": 9.777,
+      "eval_prompt_dialogue_steps_per_second": 0.816,
+      "step": 2500
+    }
+  ],
+  "max_steps": 2844,
+  "num_train_epochs": 3,
+  "total_flos": 147144228831232.0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb9607fb0053bbe67a5f7e6792f0aaef0aa879aef697c0b5c8814f6482f93625
+size 4719

zero_to_fp32.py

@@ -0,0 +1,482 @@
+#!/usr/bin/env python
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION,
+                                            OPTIMIZER_STATE_DICT,
+                                            SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS,
+                                            ZERO_STAGE,
+                                            PARTITION_COUNT,
+                                            PARAM_SHAPES,
+                                            BUFFER_NAMES)
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_optim_files(checkpoint_dir):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    optim_files = sorted(glob.glob(os.path.join(checkpoint_dir,
+                                                "*_optim_states.pt")),
+                         key=natural_keys)
+
+    if len(optim_files) == 0:
+        raise FileNotFoundError(
+            f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
+
+    return optim_files
+
+
+def parse_model_state(file):
+    state_dict = torch.load(file, map_location=device)
+
+    if BUFFER_NAMES not in state_dict:
+        raise ValueError(f"{file} is not a model state checkpoint")
+    buffer_names = state_dict[BUFFER_NAMES]
+    if debug:
+        print("Found buffers:", buffer_names)
+
+    # recover just the buffers while restoring them to fp32 if they were saved in fp16
+    buffers = {
+        k: v.float()
+        for k,
+        v in state_dict["module"].items() if k in buffer_names
+    }
+    param_shapes = state_dict[PARAM_SHAPES]
+
+    ds_version = state_dict.get(DS_VERSION, None)
+
+    return buffers, param_shapes, ds_version
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [
+            state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
+            for i in range(len(state_dicts))
+        ]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
+                      0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(
+        f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers, param_shapes, ds_version = parse_model_state(model_file)
+    print(f'Parsing checkpoint created by deepspeed=={ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(
+                    f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum([
+        full_single_fp32_vector.numel()
+        for full_single_fp32_vector in merged_single_partition_of_fp32_groups
+    ])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum(
+            [sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(
+                    f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
+                )
+            state_dict[name] = full_single_fp32_vector.narrow(
+                0,
+                offset,
+                unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(
+                f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0,
+                                             offset,
+                                             partitioned_numel)
+                  for i in range(world_size)),
+            0).narrow(0,
+                      0,
+                      unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(
+            f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "checkpoint_dir",
+        type=str,
+        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help=
+        "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
+    )
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)