Upload 8 files

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": 50004,
+  "word_embed_proj_dim": 2048
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ce076ed29967ad92e44f5866f1e11c519d72ad0277f3192f737a480201f512b
+size 2630541561

special_tokens_map.json

@@ -0,0 +1,10 @@
+{
+  "additional_special_tokens": [
+    "<bot>",
+    "<human>",
+    "</prefix>",
+    "<prefix>"
+  ],
+  "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 +1,1720 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.217287974468663,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.666666666666667e-06,
+      "loss": 2.5689,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.410009602540877e-06,
+      "loss": 2.2872,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.899727994397217e-06,
+      "loss": 2.2045,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 4.242838952070665e-06,
+      "loss": 2.1665,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.507189546742703e-06,
+      "loss": 2.0691,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.722272031045718e-06,
+      "loss": 2.0554,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.9035975752993475e-06,
+      "loss": 2.0278,
+      "step": 70
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.060338910107844e-06,
+      "loss": 2.0495,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.1983733510531e-06,
+      "loss": 2.0476,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.3216938522601335e-06,
+      "loss": 2.0547,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.433137327841662e-06,
+      "loss": 1.992,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.534791897046749e-06,
+      "loss": 2.0016,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.6282392274646635e-06,
+      "loss": 2.0251,
+      "step": 130
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.71470613401092e-06,
+      "loss": 1.9642,
+      "step": 140
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.795163382432731e-06,
+      "loss": 1.9489,
+      "step": 150
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.870392331521204e-06,
+      "loss": 1.9492,
+      "step": 160
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.941031212303129e-06,
+      "loss": 1.9794,
+      "step": 170
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 6.007608082613048e-06,
+      "loss": 1.9443,
+      "step": 180
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 6.070564811128651e-06,
+      "loss": 1.9561,
+      "step": 190
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 6.130274870425884e-06,
+      "loss": 1.9445,
+      "step": 200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 6.187056762962811e-06,
+      "loss": 1.9922,
+      "step": 210
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 6.241184306240316e-06,
+      "loss": 1.9092,
+      "step": 220
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 6.292894619573035e-06,
+      "loss": 1.9877,
+      "step": 230
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.3423944025283676e-06,
+      "loss": 1.9013,
+      "step": 240
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.385204482203244e-06,
+      "loss": 1.9351,
+      "step": 250
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.430985882568613e-06,
+      "loss": 1.9364,
+      "step": 260
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.475026117410103e-06,
+      "loss": 1.9298,
+      "step": 270
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.51745278911487e-06,
+      "loss": 1.9262,
+      "step": 280
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.55837996735795e-06,
+      "loss": 1.9245,
+      "step": 290
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.597910037536681e-06,
+      "loss": 1.9264,
+      "step": 300
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.6361352440011835e-06,
+      "loss": 1.9318,
+      "step": 310
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.673138986625154e-06,
+      "loss": 1.8992,
+      "step": 320
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.708996916564478e-06,
+      "loss": 1.8958,
+      "step": 330
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.74377786740708e-06,
+      "loss": 1.9317,
+      "step": 340
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.777544650524216e-06,
+      "loss": 1.8891,
+      "step": 350
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.810354737716999e-06,
+      "loss": 1.8944,
+      "step": 360
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.842260849796047e-06,
+      "loss": 1.8904,
+      "step": 370
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.873311466232601e-06,
+      "loss": 1.9326,
+      "step": 380
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.903551268251219e-06,
+      "loss": 1.8795,
+      "step": 390
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.933021525529835e-06,
+      "loss": 1.8729,
+      "step": 400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.961760434906346e-06,
+      "loss": 1.861,
+      "step": 410
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 6.9898034180667605e-06,
+      "loss": 1.9003,
+      "step": 420
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.017183384035125e-06,
+      "loss": 1.9061,
+      "step": 430
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.043930961344266e-06,
+      "loss": 1.9146,
+      "step": 440
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.07007470399505e-06,
+      "loss": 1.8741,
+      "step": 450
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.095641274676984e-06,
+      "loss": 1.8402,
+      "step": 460
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.120655608197665e-06,
+      "loss": 1.9109,
+      "step": 470
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.145141057632318e-06,
+      "loss": 1.8829,
+      "step": 480
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.169119525340562e-06,
+      "loss": 1.8557,
+      "step": 490
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 7.19261158069258e-06,
+      "loss": 1.8577,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_webgpt_accuracy": 0.4847224518236187,
+      "eval_webgpt_loss": 2.392578125,
+      "eval_webgpt_runtime": 337.627,
+      "eval_webgpt_samples_per_second": 11.599,
+      "eval_webgpt_steps_per_second": 0.726,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_prompt_dialogue_accuracy": 0.5612746203747981,
+      "eval_prompt_dialogue_loss": 1.828125,
+      "eval_prompt_dialogue_runtime": 811.1162,
+      "eval_prompt_dialogue_samples_per_second": 12.71,
+      "eval_prompt_dialogue_steps_per_second": 0.795,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_squad_v2_accuracy": 0.9115060111379335,
+      "eval_squad_v2_loss": NaN,
+      "eval_squad_v2_runtime": 941.3453,
+      "eval_squad_v2_samples_per_second": 12.613,
+      "eval_squad_v2_steps_per_second": 0.789,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_adversarial_qa_accuracy": 0.8083160800552105,
+      "eval_adversarial_qa_loss": 0.86083984375,
+      "eval_adversarial_qa_runtime": 238.8648,
+      "eval_adversarial_qa_samples_per_second": 12.559,
+      "eval_adversarial_qa_steps_per_second": 0.787,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_trivia_qa_nocontext_accuracy": 0.45998188422713815,
+      "eval_trivia_qa_nocontext_loss": 3.12890625,
+      "eval_trivia_qa_nocontext_runtime": 1249.2124,
+      "eval_trivia_qa_nocontext_samples_per_second": 14.364,
+      "eval_trivia_qa_nocontext_steps_per_second": 0.898,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_m2m_translation_accuracy": 0.5733300105250808,
+      "eval_m2m_translation_loss": 1.8857421875,
+      "eval_m2m_translation_runtime": 3725.0595,
+      "eval_m2m_translation_samples_per_second": 14.208,
+      "eval_m2m_translation_steps_per_second": 0.888,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_xsum_accuracy": 0.5671291686677834,
+      "eval_xsum_loss": 1.9658203125,
+      "eval_xsum_runtime": 1011.0399,
+      "eval_xsum_samples_per_second": 11.208,
+      "eval_xsum_steps_per_second": 0.701,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_cnn_dailymail_accuracy": 0.6565405592058271,
+      "eval_cnn_dailymail_loss": 1.5810546875,
+      "eval_cnn_dailymail_runtime": 1213.6108,
+      "eval_cnn_dailymail_samples_per_second": 11.015,
+      "eval_cnn_dailymail_steps_per_second": 0.689,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_multi_news_accuracy": 0.5175576159426271,
+      "eval_multi_news_loss": 2.33203125,
+      "eval_multi_news_runtime": 512.3687,
+      "eval_multi_news_samples_per_second": 10.973,
+      "eval_multi_news_steps_per_second": 0.687,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_scitldr_accuracy": 0.46241087965899574,
+      "eval_scitldr_loss": 2.5234375,
+      "eval_scitldr_runtime": 57.2333,
+      "eval_scitldr_samples_per_second": 10.815,
+      "eval_scitldr_steps_per_second": 0.681,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_soda_accuracy": 0.7121364530942489,
+      "eval_soda_loss": 1.21484375,
+      "eval_soda_runtime": 82.3256,
+      "eval_soda_samples_per_second": 12.159,
+      "eval_soda_steps_per_second": 0.765,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_joke_accuracy": 0.454450170320552,
+      "eval_joke_loss": 2.62109375,
+      "eval_joke_runtime": 6.9296,
+      "eval_joke_samples_per_second": 10.967,
+      "eval_joke_steps_per_second": 0.722,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_gsm8k_accuracy": 0.8305793220560381,
+      "eval_gsm8k_loss": 0.61328125,
+      "eval_gsm8k_runtime": 113.5097,
+      "eval_gsm8k_samples_per_second": 11.62,
+      "eval_gsm8k_steps_per_second": 0.731,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_samsum_accuracy": 0.6179719084733047,
+      "eval_samsum_loss": 1.5537109375,
+      "eval_samsum_runtime": 73.3198,
+      "eval_samsum_samples_per_second": 11.157,
+      "eval_samsum_steps_per_second": 0.709,
+      "step": 500
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.215636566090451e-06,
+      "loss": 1.8655,
+      "step": 510
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.238212692653954e-06,
+      "loss": 1.9284,
+      "step": 520
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.260357126756832e-06,
+      "loss": 1.8522,
+      "step": 530
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.282086068443704e-06,
+      "loss": 1.8869,
+      "step": 540
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.303414822624985e-06,
+      "loss": 1.905,
+      "step": 550
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.324357863833543e-06,
+      "loss": 1.8483,
+      "step": 560
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.344928895229384e-06,
+      "loss": 1.8628,
+      "step": 570
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.365140902454744e-06,
+      "loss": 1.8982,
+      "step": 580
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.385006202869703e-06,
+      "loss": 1.834,
+      "step": 590
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.404536490635762e-06,
+      "loss": 1.8488,
+      "step": 600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.423742878060626e-06,
+      "loss": 1.8585,
+      "step": 610
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.442635933570175e-06,
+      "loss": 1.8292,
+      "step": 620
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.461225716632522e-06,
+      "loss": 1.8399,
+      "step": 630
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.479521809923099e-06,
+      "loss": 1.8775,
+      "step": 640
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.497533348988249e-06,
+      "loss": 1.8593,
+      "step": 650
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.515269049637215e-06,
+      "loss": 1.8557,
+      "step": 660
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.532737233268121e-06,
+      "loss": 1.8427,
+      "step": 670
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.5499458503121685e-06,
+      "loss": 1.859,
+      "step": 680
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 7.566902501961364e-06,
+      "loss": 1.8111,
+      "step": 690
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.5836144603284295e-06,
+      "loss": 1.8246,
+      "step": 700
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.600088687172717e-06,
+      "loss": 1.8613,
+      "step": 710
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.616331851312801e-06,
+      "loss": 1.8478,
+      "step": 720
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.632350344834765e-06,
+      "loss": 1.8839,
+      "step": 730
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.648150298194777e-06,
+      "loss": 1.8625,
+      "step": 740
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.663737594305231e-06,
+      "loss": 1.8291,
+      "step": 750
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.679117881685476e-06,
+      "loss": 1.836,
+      "step": 760
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.6942965867507e-06,
+      "loss": 1.8102,
+      "step": 770
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.709278925305836e-06,
+      "loss": 1.8353,
+      "step": 780
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.72406991330548e-06,
+      "loss": 1.8713,
+      "step": 790
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.738674376935279e-06,
+      "loss": 1.8314,
+      "step": 800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.753096962065562e-06,
+      "loss": 1.8553,
+      "step": 810
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.767342143123527e-06,
+      "loss": 1.8318,
+      "step": 820
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.781414231426347e-06,
+      "loss": 1.8401,
+      "step": 830
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.79531738301407e-06,
+      "loss": 1.8108,
+      "step": 840
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.809055606017904e-06,
+      "loss": 1.8563,
+      "step": 850
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.822632767596549e-06,
+      "loss": 1.8408,
+      "step": 860
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.836052600470646e-06,
+      "loss": 1.8096,
+      "step": 870
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.849318709082941e-06,
+      "loss": 1.8264,
+      "step": 880
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.862434575409602e-06,
+      "loss": 1.8902,
+      "step": 890
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.875403564446145e-06,
+      "loss": 1.7953,
+      "step": 900
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.888228929389561e-06,
+      "loss": 1.837,
+      "step": 910
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.900913816536647e-06,
+      "loss": 1.7704,
+      "step": 920
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.913461269916965e-06,
+      "loss": 1.8251,
+      "step": 930
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.925874235677506e-06,
+      "loss": 1.8294,
+      "step": 940
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.938155566234842e-06,
+      "loss": 1.8306,
+      "step": 950
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.950308024209451e-06,
+      "loss": 1.7889,
+      "step": 960
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.962334286155715e-06,
+      "loss": 1.7922,
+      "step": 970
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.974236946100272e-06,
+      "loss": 1.7679,
+      "step": 980
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.986018518900343e-06,
+      "loss": 1.8068,
+      "step": 990
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.997681443432989e-06,
+      "loss": 1.8467,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_webgpt_accuracy": 0.4853852488269729,
+      "eval_webgpt_loss": 2.388671875,
+      "eval_webgpt_runtime": 337.66,
+      "eval_webgpt_samples_per_second": 11.597,
+      "eval_webgpt_steps_per_second": 0.726,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_prompt_dialogue_accuracy": 0.5656945584798752,
+      "eval_prompt_dialogue_loss": 1.7978515625,
+      "eval_prompt_dialogue_runtime": 810.7166,
+      "eval_prompt_dialogue_samples_per_second": 12.716,
+      "eval_prompt_dialogue_steps_per_second": 0.796,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_squad_v2_accuracy": 0.929180907765482,
+      "eval_squad_v2_loss": NaN,
+      "eval_squad_v2_runtime": 942.6897,
+      "eval_squad_v2_samples_per_second": 12.595,
+      "eval_squad_v2_steps_per_second": 0.788,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_adversarial_qa_accuracy": 0.8106740280653324,
+      "eval_adversarial_qa_loss": 0.8291015625,
+      "eval_adversarial_qa_runtime": 239.9039,
+      "eval_adversarial_qa_samples_per_second": 12.505,
+      "eval_adversarial_qa_steps_per_second": 0.784,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_trivia_qa_nocontext_accuracy": 0.46837345154352583,
+      "eval_trivia_qa_nocontext_loss": 3.052734375,
+      "eval_trivia_qa_nocontext_runtime": 1248.9886,
+      "eval_trivia_qa_nocontext_samples_per_second": 14.367,
+      "eval_trivia_qa_nocontext_steps_per_second": 0.898,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_m2m_translation_accuracy": 0.6128791362753888,
+      "eval_m2m_translation_loss": 1.6904296875,
+      "eval_m2m_translation_runtime": 3724.3046,
+      "eval_m2m_translation_samples_per_second": 14.21,
+      "eval_m2m_translation_steps_per_second": 0.888,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_xsum_accuracy": 0.576201200101912,
+      "eval_xsum_loss": 1.90625,
+      "eval_xsum_runtime": 1013.837,
+      "eval_xsum_samples_per_second": 11.177,
+      "eval_xsum_steps_per_second": 0.699,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_cnn_dailymail_accuracy": 0.6590085257947359,
+      "eval_cnn_dailymail_loss": 1.5537109375,
+      "eval_cnn_dailymail_runtime": 1212.8269,
+      "eval_cnn_dailymail_samples_per_second": 11.022,
+      "eval_cnn_dailymail_steps_per_second": 0.689,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_multi_news_accuracy": 0.5213214574389622,
+      "eval_multi_news_loss": 2.302734375,
+      "eval_multi_news_runtime": 514.4889,
+      "eval_multi_news_samples_per_second": 10.927,
+      "eval_multi_news_steps_per_second": 0.684,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_scitldr_accuracy": 0.4667123357182727,
+      "eval_scitldr_loss": 2.50390625,
+      "eval_scitldr_runtime": 55.8523,
+      "eval_scitldr_samples_per_second": 11.083,
+      "eval_scitldr_steps_per_second": 0.698,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_soda_accuracy": 0.7221954952984911,
+      "eval_soda_loss": 1.1513671875,
+      "eval_soda_runtime": 82.1743,
+      "eval_soda_samples_per_second": 12.181,
+      "eval_soda_steps_per_second": 0.767,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_joke_accuracy": 0.45549829679447984,
+      "eval_joke_loss": 2.5625,
+      "eval_joke_runtime": 7.1872,
+      "eval_joke_samples_per_second": 10.574,
+      "eval_joke_steps_per_second": 0.696,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_gsm8k_accuracy": 0.836947292796029,
+      "eval_gsm8k_loss": 0.591796875,
+      "eval_gsm8k_runtime": 112.7859,
+      "eval_gsm8k_samples_per_second": 11.695,
+      "eval_gsm8k_steps_per_second": 0.736,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_samsum_accuracy": 0.6263182026593306,
+      "eval_samsum_loss": 1.5126953125,
+      "eval_samsum_runtime": 73.6542,
+      "eval_samsum_samples_per_second": 11.106,
+      "eval_samsum_steps_per_second": 0.706,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.997895686156621e-06,
+      "loss": 1.8114,
+      "step": 1010
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.994889523523223e-06,
+      "loss": 1.857,
+      "step": 1020
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.991883360889824e-06,
+      "loss": 1.8195,
+      "step": 1030
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.988877198256426e-06,
+      "loss": 1.8101,
+      "step": 1040
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.985871035623028e-06,
+      "loss": 1.8169,
+      "step": 1050
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.982864872989627e-06,
+      "loss": 1.8414,
+      "step": 1060
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.979858710356229e-06,
+      "loss": 1.825,
+      "step": 1070
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.97685254772283e-06,
+      "loss": 1.7882,
+      "step": 1080
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.973846385089432e-06,
+      "loss": 1.7857,
+      "step": 1090
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.970840222456035e-06,
+      "loss": 1.8229,
+      "step": 1100
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.967834059822637e-06,
+      "loss": 1.8152,
+      "step": 1110
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.964827897189238e-06,
+      "loss": 1.7865,
+      "step": 1120
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.96182173455584e-06,
+      "loss": 1.8038,
+      "step": 1130
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.958815571922442e-06,
+      "loss": 1.796,
+      "step": 1140
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.955809409289041e-06,
+      "loss": 1.7737,
+      "step": 1150
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.952803246655643e-06,
+      "loss": 1.7958,
+      "step": 1160
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.949797084022244e-06,
+      "loss": 1.7981,
+      "step": 1170
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.946790921388846e-06,
+      "loss": 1.8061,
+      "step": 1180
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.943784758755448e-06,
+      "loss": 1.8395,
+      "step": 1190
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.940778596122049e-06,
+      "loss": 1.7803,
+      "step": 1200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.937772433488652e-06,
+      "loss": 1.7906,
+      "step": 1210
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.934766270855254e-06,
+      "loss": 1.8076,
+      "step": 1220
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.931760108221855e-06,
+      "loss": 1.7971,
+      "step": 1230
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.928753945588455e-06,
+      "loss": 1.8292,
+      "step": 1240
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.925747782955057e-06,
+      "loss": 1.7876,
+      "step": 1250
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.922741620321658e-06,
+      "loss": 1.7499,
+      "step": 1260
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.91973545768826e-06,
+      "loss": 1.8209,
+      "step": 1270
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.916729295054861e-06,
+      "loss": 1.7971,
+      "step": 1280
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.913723132421463e-06,
+      "loss": 1.8168,
+      "step": 1290
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.910716969788065e-06,
+      "loss": 1.7771,
+      "step": 1300
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.907710807154668e-06,
+      "loss": 1.7611,
+      "step": 1310
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.90470464452127e-06,
+      "loss": 1.7673,
+      "step": 1320
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.90169848188787e-06,
+      "loss": 1.8694,
+      "step": 1330
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.89869231925447e-06,
+      "loss": 1.7543,
+      "step": 1340
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.895686156621072e-06,
+      "loss": 1.759,
+      "step": 1350
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.892679993987674e-06,
+      "loss": 1.7929,
+      "step": 1360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.889673831354275e-06,
+      "loss": 1.8223,
+      "step": 1370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.886667668720877e-06,
+      "loss": 1.7829,
+      "step": 1380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.883661506087479e-06,
+      "loss": 1.8088,
+      "step": 1390
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.88065534345408e-06,
+      "loss": 1.8012,
+      "step": 1400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.877649180820682e-06,
+      "loss": 1.7337,
+      "step": 1410
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.874643018187283e-06,
+      "loss": 1.7848,
+      "step": 1420
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.871636855553885e-06,
+      "loss": 1.7741,
+      "step": 1430
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.868630692920486e-06,
+      "loss": 1.8111,
+      "step": 1440
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.865624530287088e-06,
+      "loss": 1.7663,
+      "step": 1450
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.86261836765369e-06,
+      "loss": 1.7635,
+      "step": 1460
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.859612205020291e-06,
+      "loss": 1.7613,
+      "step": 1470
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.856606042386892e-06,
+      "loss": 1.7537,
+      "step": 1480
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.853599879753494e-06,
+      "loss": 1.8094,
+      "step": 1490
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.850593717120096e-06,
+      "loss": 1.7487,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_webgpt_accuracy": 0.486666038374247,
+      "eval_webgpt_loss": 2.384765625,
+      "eval_webgpt_runtime": 336.8175,
+      "eval_webgpt_samples_per_second": 11.626,
+      "eval_webgpt_steps_per_second": 0.727,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_prompt_dialogue_accuracy": 0.5696171266680501,
+      "eval_prompt_dialogue_loss": 1.7705078125,
+      "eval_prompt_dialogue_runtime": 809.6613,
+      "eval_prompt_dialogue_samples_per_second": 12.732,
+      "eval_prompt_dialogue_steps_per_second": 0.797,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_squad_v2_accuracy": 0.9273763713281866,
+      "eval_squad_v2_loss": NaN,
+      "eval_squad_v2_runtime": 941.2422,
+      "eval_squad_v2_samples_per_second": 12.614,
+      "eval_squad_v2_steps_per_second": 0.789,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_adversarial_qa_accuracy": 0.8228088336783989,
+      "eval_adversarial_qa_loss": 0.76416015625,
+      "eval_adversarial_qa_runtime": 238.4374,
+      "eval_adversarial_qa_samples_per_second": 12.582,
+      "eval_adversarial_qa_steps_per_second": 0.788,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_trivia_qa_nocontext_accuracy": 0.47413712205101255,
+      "eval_trivia_qa_nocontext_loss": 2.9921875,
+      "eval_trivia_qa_nocontext_runtime": 1248.2736,
+      "eval_trivia_qa_nocontext_samples_per_second": 14.375,
+      "eval_trivia_qa_nocontext_steps_per_second": 0.899,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_m2m_translation_accuracy": 0.6384202273383193,
+      "eval_m2m_translation_loss": 1.5703125,
+      "eval_m2m_translation_runtime": 3725.8719,
+      "eval_m2m_translation_samples_per_second": 14.204,
+      "eval_m2m_translation_steps_per_second": 0.888,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_xsum_accuracy": 0.5815857676257413,
+      "eval_xsum_loss": 1.87109375,
+      "eval_xsum_runtime": 1009.1324,
+      "eval_xsum_samples_per_second": 11.229,
+      "eval_xsum_steps_per_second": 0.703,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_cnn_dailymail_accuracy": 0.6627513117154793,
+      "eval_cnn_dailymail_loss": 1.537109375,
+      "eval_cnn_dailymail_runtime": 1211.2046,
+      "eval_cnn_dailymail_samples_per_second": 11.037,
+      "eval_cnn_dailymail_steps_per_second": 0.69,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_multi_news_accuracy": 0.5242124214780086,
+      "eval_multi_news_loss": 2.287109375,
+      "eval_multi_news_runtime": 513.1245,
+      "eval_multi_news_samples_per_second": 10.956,
+      "eval_multi_news_steps_per_second": 0.686,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_scitldr_accuracy": 0.46804191723122984,
+      "eval_scitldr_loss": 2.4921875,
+      "eval_scitldr_runtime": 56.2815,
+      "eval_scitldr_samples_per_second": 10.998,
+      "eval_scitldr_steps_per_second": 0.693,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_soda_accuracy": 0.7266564618412421,
+      "eval_soda_loss": 1.12890625,
+      "eval_soda_runtime": 82.0556,
+      "eval_soda_samples_per_second": 12.199,
+      "eval_soda_steps_per_second": 0.768,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_joke_accuracy": 0.45995283430867323,
+      "eval_joke_loss": 2.5,
+      "eval_joke_runtime": 7.0053,
+      "eval_joke_samples_per_second": 10.849,
+      "eval_joke_steps_per_second": 0.714,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_gsm8k_accuracy": 0.8421233100385344,
+      "eval_gsm8k_loss": 0.57177734375,
+      "eval_gsm8k_runtime": 113.8436,
+      "eval_gsm8k_samples_per_second": 11.586,
+      "eval_gsm8k_steps_per_second": 0.729,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_samsum_accuracy": 0.6305018339446482,
+      "eval_samsum_loss": 1.490234375,
+      "eval_samsum_runtime": 72.6647,
+      "eval_samsum_samples_per_second": 11.257,
+      "eval_samsum_steps_per_second": 0.716,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.847587554486697e-06,
+      "loss": 1.7693,
+      "step": 1510
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.844882008116638e-06,
+      "loss": 1.7719,
+      "step": 1520
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.84187584548324e-06,
+      "loss": 1.7777,
+      "step": 1530
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.838869682849841e-06,
+      "loss": 1.7922,
+      "step": 1540
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.835863520216443e-06,
+      "loss": 1.7688,
+      "step": 1550
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.832857357583046e-06,
+      "loss": 1.7848,
+      "step": 1560
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.829851194949648e-06,
+      "loss": 1.7795,
+      "step": 1570
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.826845032316247e-06,
+      "loss": 1.7826,
+      "step": 1580
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.823838869682849e-06,
+      "loss": 1.7511,
+      "step": 1590
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.82083270704945e-06,
+      "loss": 1.7502,
+      "step": 1600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.817826544416052e-06,
+      "loss": 1.7384,
+      "step": 1610
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.814820381782654e-06,
+      "loss": 1.773,
+      "step": 1620
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.811814219149255e-06,
+      "loss": 1.785,
+      "step": 1630
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.808808056515857e-06,
+      "loss": 1.7407,
+      "step": 1640
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.805801893882458e-06,
+      "loss": 1.8115,
+      "step": 1650
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.80279573124906e-06,
+      "loss": 1.8207,
+      "step": 1660
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.799789568615661e-06,
+      "loss": 1.7618,
+      "step": 1670
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.796783405982263e-06,
+      "loss": 1.7725,
+      "step": 1680
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.793777243348864e-06,
+      "loss": 1.7618,
+      "step": 1690
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.790771080715466e-06,
+      "loss": 1.7567,
+      "step": 1700
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.787764918082068e-06,
+      "loss": 1.7704,
+      "step": 1710
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.784758755448669e-06,
+      "loss": 1.7426,
+      "step": 1720
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.78175259281527e-06,
+      "loss": 1.7373,
+      "step": 1730
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.778746430181872e-06,
+      "loss": 1.7966,
+      "step": 1740
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.775740267548474e-06,
+      "loss": 1.773,
+      "step": 1750
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.772734104915075e-06,
+      "loss": 1.7664,
+      "step": 1760
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.769727942281677e-06,
+      "loss": 1.7518,
+      "step": 1770
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.766721779648278e-06,
+      "loss": 1.7743,
+      "step": 1780
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.76371561701488e-06,
+      "loss": 1.7882,
+      "step": 1790
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.760709454381481e-06,
+      "loss": 1.7534,
+      "step": 1800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.757703291748083e-06,
+      "loss": 1.7761,
+      "step": 1810
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.754697129114685e-06,
+      "loss": 1.7951,
+      "step": 1820
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.751690966481286e-06,
+      "loss": 1.7287,
+      "step": 1830
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.748684803847888e-06,
+      "loss": 1.7674,
+      "step": 1840
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.74567864121449e-06,
+      "loss": 1.7664,
+      "step": 1850
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.74267247858109e-06,
+      "loss": 1.7569,
+      "step": 1860
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.739666315947692e-06,
+      "loss": 1.8026,
+      "step": 1870
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.736960769577633e-06,
+      "loss": 1.7262,
+      "step": 1880
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.733954606944235e-06,
+      "loss": 1.7266,
+      "step": 1890
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.730948444310836e-06,
+      "loss": 1.7476,
+      "step": 1900
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.727942281677438e-06,
+      "loss": 1.7544,
+      "step": 1910
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.72493611904404e-06,
+      "loss": 1.7859,
+      "step": 1920
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.721929956410641e-06,
+      "loss": 1.7386,
+      "step": 1930
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.718923793777243e-06,
+      "loss": 1.7012,
+      "step": 1940
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.715917631143844e-06,
+      "loss": 1.7736,
+      "step": 1950
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.712911468510446e-06,
+      "loss": 1.7398,
+      "step": 1960
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.709905305877047e-06,
+      "loss": 1.8026,
+      "step": 1970
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.706899143243649e-06,
+      "loss": 1.798,
+      "step": 1980
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.70389298061025e-06,
+      "loss": 1.802,
+      "step": 1990
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.700886817976852e-06,
+      "loss": 1.7447,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_webgpt_accuracy": 0.48722841158921415,
+      "eval_webgpt_loss": 2.37890625,
+      "eval_webgpt_runtime": 335.1847,
+      "eval_webgpt_samples_per_second": 11.683,
+      "eval_webgpt_steps_per_second": 0.731,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_prompt_dialogue_accuracy": 0.5729272869993594,
+      "eval_prompt_dialogue_loss": 1.7509765625,
+      "eval_prompt_dialogue_runtime": 808.5089,
+      "eval_prompt_dialogue_samples_per_second": 12.751,
+      "eval_prompt_dialogue_steps_per_second": 0.798,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_squad_v2_accuracy": 0.9350127871125027,
+      "eval_squad_v2_loss": NaN,
+      "eval_squad_v2_runtime": 937.9399,
+      "eval_squad_v2_samples_per_second": 12.659,
+      "eval_squad_v2_steps_per_second": 0.792,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_adversarial_qa_accuracy": 0.8308028525419829,
+      "eval_adversarial_qa_loss": 0.7265625,
+      "eval_adversarial_qa_runtime": 238.5652,
+      "eval_adversarial_qa_samples_per_second": 12.575,
+      "eval_adversarial_qa_steps_per_second": 0.788,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_trivia_qa_nocontext_accuracy": 0.47783426824006153,
+      "eval_trivia_qa_nocontext_loss": 2.9609375,
+      "eval_trivia_qa_nocontext_runtime": 1249.121,
+      "eval_trivia_qa_nocontext_samples_per_second": 14.365,
+      "eval_trivia_qa_nocontext_steps_per_second": 0.898,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_m2m_translation_accuracy": 0.6576231472988909,
+      "eval_m2m_translation_loss": 1.4755859375,
+      "eval_m2m_translation_runtime": 3725.504,
+      "eval_m2m_translation_samples_per_second": 14.206,
+      "eval_m2m_translation_steps_per_second": 0.888,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_xsum_accuracy": 0.5862788967827711,
+      "eval_xsum_loss": 1.8447265625,
+      "eval_xsum_runtime": 1007.4241,
+      "eval_xsum_samples_per_second": 11.248,
+      "eval_xsum_steps_per_second": 0.704,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_cnn_dailymail_accuracy": 0.6635118400214548,
+      "eval_cnn_dailymail_loss": 1.5224609375,
+      "eval_cnn_dailymail_runtime": 1207.7216,
+      "eval_cnn_dailymail_samples_per_second": 11.069,
+      "eval_cnn_dailymail_steps_per_second": 0.692,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_multi_news_accuracy": 0.5254140019732162,
+      "eval_multi_news_loss": 2.271484375,
+      "eval_multi_news_runtime": 511.4499,
+      "eval_multi_news_samples_per_second": 10.992,
+      "eval_multi_news_steps_per_second": 0.688,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_scitldr_accuracy": 0.46584969807682547,
+      "eval_scitldr_loss": 2.50390625,
+      "eval_scitldr_runtime": 56.1361,
+      "eval_scitldr_samples_per_second": 11.027,
+      "eval_scitldr_steps_per_second": 0.695,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_soda_accuracy": 0.7280997157227204,
+      "eval_soda_loss": 1.1171875,
+      "eval_soda_runtime": 81.9632,
+      "eval_soda_samples_per_second": 12.213,
+      "eval_soda_steps_per_second": 0.769,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_joke_accuracy": 0.4692986287011966,
+      "eval_joke_loss": 2.46484375,
+      "eval_joke_runtime": 6.7476,
+      "eval_joke_samples_per_second": 11.263,
+      "eval_joke_steps_per_second": 0.741,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_gsm8k_accuracy": 0.8437234667885833,
+      "eval_gsm8k_loss": 0.5634765625,
+      "eval_gsm8k_runtime": 113.7256,
+      "eval_gsm8k_samples_per_second": 11.598,
+      "eval_gsm8k_steps_per_second": 0.73,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_samsum_accuracy": 0.6325025008336113,
+      "eval_samsum_loss": 1.4638671875,
+      "eval_samsum_runtime": 72.3462,
+      "eval_samsum_samples_per_second": 11.307,
+      "eval_samsum_steps_per_second": 0.719,
+      "step": 2000
+    }
+  ],
+  "max_steps": 27612,
+  "num_train_epochs": 3,
+  "total_flos": 428521886187520.0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:250f3ddd3f36842fe2e5d53c77d84a300a9acbd55b30c6b38e058f6650458b60
+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)