First model version

added_tokens.json

@@ -0,0 +1,3 @@
+{
+  "[PAD]": 32000
+}

config.json

@@ -0,0 +1,23 @@
+{
+  "_name_or_path": "../../../PLMs/ft_local/LLAMA/hf/llama-7b",
+  "architectures": [
+    "LLaMAForCausalLM"
+  ],
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_sequence_length": 2048,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "pad_token_id": 0,
+  "rms_norm_eps": 1e-06,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.27.0.dev0",
+  "use_cache": true,
+  "vocab_size": 32001
+}

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "pad_token_id": 0,
+  "transformers_version": "4.27.0.dev0"
+}

latest

@@ -0,0 +1 @@
+global_step375

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45f76a5eb2049a4310186988aa8792bf1e7e8ff68f47355d62f9084364c1201e
+size 26953810889

special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "</s>",
+  "eos_token": "</s>",
+  "pad_token": "[PAD]",
+  "unk_token": "</s>"
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,9 @@
+{
+  "bos_token": "",
+  "eos_token": "",
+  "model_max_length": 512,
+  "padding_side": "right",
+  "special_tokens_map_file": "../../../PLMs/ft_local/LLAMA/hf/llama-7b/special_tokens_map.json",
+  "tokenizer_class": "LLaMATokenizer",
+  "unk_token": ""
+}

trainer_state.json

@@ -0,0 +1,2275 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "global_step": 375,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0,
+      "loss": 1.4456,
+      "step": 1
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.578858913022597e-06,
+      "loss": 1.2331,
+      "step": 2
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 8.842282173954808e-06,
+      "loss": 1.2688,
+      "step": 3
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 1.1157717826045193e-05,
+      "loss": 1.2466,
+      "step": 4
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 1.295370924755994e-05,
+      "loss": 1.2583,
+      "step": 5
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.4421141086977404e-05,
+      "loss": 1.1955,
+      "step": 6
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.5661837028938922e-05,
+      "loss": 1.1408,
+      "step": 7
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.6736576739067793e-05,
+      "loss": 1.2148,
+      "step": 8
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.7684564347909616e-05,
+      "loss": 1.1095,
+      "step": 9
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1.853256816058254e-05,
+      "loss": 1.2244,
+      "step": 10
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1.929968091962688e-05,
+      "loss": 1.1805,
+      "step": 11
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 2e-05,
+      "loss": 1.1408,
+      "step": 12
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 2e-05,
+      "loss": 1.2106,
+      "step": 13
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1.994490358126722e-05,
+      "loss": 1.1368,
+      "step": 14
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9889807162534438e-05,
+      "loss": 1.1637,
+      "step": 15
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.9834710743801656e-05,
+      "loss": 1.1587,
+      "step": 16
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.977961432506887e-05,
+      "loss": 1.1201,
+      "step": 17
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.972451790633609e-05,
+      "loss": 1.1684,
+      "step": 18
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1.9669421487603307e-05,
+      "loss": 1.1014,
+      "step": 19
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.9614325068870526e-05,
+      "loss": 1.1248,
+      "step": 20
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.9559228650137744e-05,
+      "loss": 1.1479,
+      "step": 21
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1.950413223140496e-05,
+      "loss": 1.1321,
+      "step": 22
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1.944903581267218e-05,
+      "loss": 1.1393,
+      "step": 23
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.9393939393939395e-05,
+      "loss": 1.1275,
+      "step": 24
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1.9338842975206613e-05,
+      "loss": 1.0823,
+      "step": 25
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.928374655647383e-05,
+      "loss": 1.1044,
+      "step": 26
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1.922865013774105e-05,
+      "loss": 1.122,
+      "step": 27
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1.9173553719008268e-05,
+      "loss": 1.1913,
+      "step": 28
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.9118457300275483e-05,
+      "loss": 1.1471,
+      "step": 29
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.90633608815427e-05,
+      "loss": 1.1275,
+      "step": 30
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1.900826446280992e-05,
+      "loss": 1.1164,
+      "step": 31
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.8953168044077137e-05,
+      "loss": 1.0628,
+      "step": 32
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.8898071625344356e-05,
+      "loss": 1.0881,
+      "step": 33
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1.884297520661157e-05,
+      "loss": 1.116,
+      "step": 34
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.8787878787878792e-05,
+      "loss": 1.1315,
+      "step": 35
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.8732782369146007e-05,
+      "loss": 1.1533,
+      "step": 36
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.8677685950413225e-05,
+      "loss": 1.1235,
+      "step": 37
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.8622589531680443e-05,
+      "loss": 1.113,
+      "step": 38
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.856749311294766e-05,
+      "loss": 1.1017,
+      "step": 39
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1.851239669421488e-05,
+      "loss": 1.0719,
+      "step": 40
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.8457300275482094e-05,
+      "loss": 1.0921,
+      "step": 41
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.8402203856749313e-05,
+      "loss": 1.1365,
+      "step": 42
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.834710743801653e-05,
+      "loss": 1.0933,
+      "step": 43
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.829201101928375e-05,
+      "loss": 1.1952,
+      "step": 44
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.8236914600550967e-05,
+      "loss": 1.0772,
+      "step": 45
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.8181818181818182e-05,
+      "loss": 1.0692,
+      "step": 46
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.8126721763085404e-05,
+      "loss": 1.0966,
+      "step": 47
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.807162534435262e-05,
+      "loss": 1.1473,
+      "step": 48
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1.8016528925619837e-05,
+      "loss": 1.063,
+      "step": 49
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.796143250688705e-05,
+      "loss": 1.0702,
+      "step": 50
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.7906336088154273e-05,
+      "loss": 1.1028,
+      "step": 51
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.7851239669421488e-05,
+      "loss": 1.1121,
+      "step": 52
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.7796143250688706e-05,
+      "loss": 1.0962,
+      "step": 53
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.7741046831955924e-05,
+      "loss": 1.1432,
+      "step": 54
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.7685950413223143e-05,
+      "loss": 1.1637,
+      "step": 55
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.763085399449036e-05,
+      "loss": 1.1187,
+      "step": 56
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.7575757575757576e-05,
+      "loss": 1.0772,
+      "step": 57
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.7520661157024794e-05,
+      "loss": 1.1177,
+      "step": 58
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.7465564738292012e-05,
+      "loss": 1.1503,
+      "step": 59
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1.741046831955923e-05,
+      "loss": 1.135,
+      "step": 60
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.735537190082645e-05,
+      "loss": 1.0959,
+      "step": 61
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.7300275482093663e-05,
+      "loss": 1.1231,
+      "step": 62
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.7245179063360885e-05,
+      "loss": 1.1052,
+      "step": 63
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.71900826446281e-05,
+      "loss": 1.0436,
+      "step": 64
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.7134986225895318e-05,
+      "loss": 1.0936,
+      "step": 65
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.7079889807162536e-05,
+      "loss": 1.1038,
+      "step": 66
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.7024793388429754e-05,
+      "loss": 1.114,
+      "step": 67
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.6969696969696972e-05,
+      "loss": 1.0471,
+      "step": 68
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.6914600550964187e-05,
+      "loss": 1.1193,
+      "step": 69
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.6859504132231405e-05,
+      "loss": 1.1197,
+      "step": 70
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.6804407713498624e-05,
+      "loss": 1.0495,
+      "step": 71
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6749311294765842e-05,
+      "loss": 1.0766,
+      "step": 72
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.669421487603306e-05,
+      "loss": 1.0873,
+      "step": 73
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6639118457300275e-05,
+      "loss": 1.1187,
+      "step": 74
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6584022038567496e-05,
+      "loss": 1.124,
+      "step": 75
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.652892561983471e-05,
+      "loss": 1.1271,
+      "step": 76
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.647382920110193e-05,
+      "loss": 1.0794,
+      "step": 77
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.6418732782369148e-05,
+      "loss": 1.0647,
+      "step": 78
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.6363636363636366e-05,
+      "loss": 1.155,
+      "step": 79
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.6308539944903584e-05,
+      "loss": 1.0537,
+      "step": 80
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.62534435261708e-05,
+      "loss": 1.1242,
+      "step": 81
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.6198347107438017e-05,
+      "loss": 1.0871,
+      "step": 82
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.6143250688705235e-05,
+      "loss": 1.0448,
+      "step": 83
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.6088154269972454e-05,
+      "loss": 1.1028,
+      "step": 84
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.6033057851239672e-05,
+      "loss": 1.1275,
+      "step": 85
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.5977961432506887e-05,
+      "loss": 1.1203,
+      "step": 86
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.5922865013774108e-05,
+      "loss": 1.0619,
+      "step": 87
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.5867768595041323e-05,
+      "loss": 1.0802,
+      "step": 88
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.581267217630854e-05,
+      "loss": 1.0682,
+      "step": 89
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.575757575757576e-05,
+      "loss": 1.1025,
+      "step": 90
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.5702479338842978e-05,
+      "loss": 1.1141,
+      "step": 91
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.5647382920110196e-05,
+      "loss": 1.0573,
+      "step": 92
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.559228650137741e-05,
+      "loss": 1.0682,
+      "step": 93
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.553719008264463e-05,
+      "loss": 1.1002,
+      "step": 94
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1.5482093663911847e-05,
+      "loss": 1.0758,
+      "step": 95
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.5426997245179065e-05,
+      "loss": 1.1524,
+      "step": 96
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.5371900826446283e-05,
+      "loss": 1.0419,
+      "step": 97
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.5316804407713498e-05,
+      "loss": 1.0896,
+      "step": 98
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.526170798898072e-05,
+      "loss": 1.0811,
+      "step": 99
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.5206611570247936e-05,
+      "loss": 1.0717,
+      "step": 100
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.5151515151515153e-05,
+      "loss": 1.1563,
+      "step": 101
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.509641873278237e-05,
+      "loss": 1.0853,
+      "step": 102
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.504132231404959e-05,
+      "loss": 1.1502,
+      "step": 103
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1.4986225895316806e-05,
+      "loss": 1.0551,
+      "step": 104
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.4931129476584022e-05,
+      "loss": 1.0128,
+      "step": 105
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.487603305785124e-05,
+      "loss": 1.0855,
+      "step": 106
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.4820936639118459e-05,
+      "loss": 1.0681,
+      "step": 107
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.4765840220385677e-05,
+      "loss": 1.0963,
+      "step": 108
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1.4710743801652893e-05,
+      "loss": 1.0618,
+      "step": 109
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.465564738292011e-05,
+      "loss": 1.1013,
+      "step": 110
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.460055096418733e-05,
+      "loss": 1.1274,
+      "step": 111
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.4545454545454546e-05,
+      "loss": 1.0618,
+      "step": 112
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.4490358126721765e-05,
+      "loss": 1.0953,
+      "step": 113
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.4435261707988981e-05,
+      "loss": 1.0739,
+      "step": 114
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1.4380165289256201e-05,
+      "loss": 1.1187,
+      "step": 115
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.4325068870523417e-05,
+      "loss": 1.1022,
+      "step": 116
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.4269972451790634e-05,
+      "loss": 1.1464,
+      "step": 117
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.4214876033057852e-05,
+      "loss": 1.0718,
+      "step": 118
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1.415977961432507e-05,
+      "loss": 1.1309,
+      "step": 119
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.4104683195592289e-05,
+      "loss": 1.124,
+      "step": 120
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.4049586776859505e-05,
+      "loss": 1.1001,
+      "step": 121
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.3994490358126722e-05,
+      "loss": 1.0716,
+      "step": 122
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.3939393939393942e-05,
+      "loss": 1.0403,
+      "step": 123
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.3884297520661158e-05,
+      "loss": 1.0735,
+      "step": 124
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.3829201101928376e-05,
+      "loss": 1.0446,
+      "step": 125
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.3774104683195593e-05,
+      "loss": 0.8614,
+      "step": 126
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.3719008264462813e-05,
+      "loss": 0.8819,
+      "step": 127
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.3663911845730029e-05,
+      "loss": 0.89,
+      "step": 128
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 1.3608815426997246e-05,
+      "loss": 0.8193,
+      "step": 129
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.3553719008264464e-05,
+      "loss": 0.8315,
+      "step": 130
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1.3498622589531682e-05,
+      "loss": 0.837,
+      "step": 131
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.34435261707989e-05,
+      "loss": 0.8391,
+      "step": 132
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.3388429752066117e-05,
+      "loss": 0.8208,
+      "step": 133
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 0.811,
+      "step": 134
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 1.3278236914600553e-05,
+      "loss": 0.8744,
+      "step": 135
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 1.322314049586777e-05,
+      "loss": 0.8584,
+      "step": 136
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1.3168044077134988e-05,
+      "loss": 0.8192,
+      "step": 137
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1.3112947658402204e-05,
+      "loss": 0.8633,
+      "step": 138
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.3057851239669424e-05,
+      "loss": 0.822,
+      "step": 139
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1.300275482093664e-05,
+      "loss": 0.8523,
+      "step": 140
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1.2947658402203857e-05,
+      "loss": 0.8124,
+      "step": 141
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.2892561983471074e-05,
+      "loss": 0.8092,
+      "step": 142
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.2837465564738294e-05,
+      "loss": 0.8262,
+      "step": 143
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 1.278236914600551e-05,
+      "loss": 0.9132,
+      "step": 144
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.2727272727272728e-05,
+      "loss": 0.8919,
+      "step": 145
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 1.2672176308539945e-05,
+      "loss": 0.8802,
+      "step": 146
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.2617079889807165e-05,
+      "loss": 0.8697,
+      "step": 147
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.2561983471074381e-05,
+      "loss": 0.919,
+      "step": 148
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.2506887052341598e-05,
+      "loss": 0.9084,
+      "step": 149
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 1.2451790633608816e-05,
+      "loss": 0.8142,
+      "step": 150
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 1.2396694214876034e-05,
+      "loss": 0.8707,
+      "step": 151
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1.2341597796143253e-05,
+      "loss": 0.8036,
+      "step": 152
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1.2286501377410469e-05,
+      "loss": 0.8466,
+      "step": 153
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.2231404958677686e-05,
+      "loss": 0.8412,
+      "step": 154
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1.2176308539944905e-05,
+      "loss": 0.9156,
+      "step": 155
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.2121212121212122e-05,
+      "loss": 0.8147,
+      "step": 156
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 1.206611570247934e-05,
+      "loss": 0.9054,
+      "step": 157
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 1.2011019283746557e-05,
+      "loss": 0.8052,
+      "step": 158
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1.1955922865013777e-05,
+      "loss": 0.8557,
+      "step": 159
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1.1900826446280993e-05,
+      "loss": 0.8617,
+      "step": 160
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1.184573002754821e-05,
+      "loss": 0.8586,
+      "step": 161
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.1790633608815428e-05,
+      "loss": 0.8359,
+      "step": 162
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.1735537190082646e-05,
+      "loss": 0.8864,
+      "step": 163
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1.1680440771349864e-05,
+      "loss": 0.8267,
+      "step": 164
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 1.162534435261708e-05,
+      "loss": 0.8364,
+      "step": 165
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1.1570247933884297e-05,
+      "loss": 0.8296,
+      "step": 166
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.1515151515151517e-05,
+      "loss": 0.8138,
+      "step": 167
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.1460055096418734e-05,
+      "loss": 0.8329,
+      "step": 168
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.1404958677685952e-05,
+      "loss": 0.8309,
+      "step": 169
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1.1349862258953168e-05,
+      "loss": 0.8328,
+      "step": 170
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1.1294765840220388e-05,
+      "loss": 0.8375,
+      "step": 171
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 1.1239669421487605e-05,
+      "loss": 0.8477,
+      "step": 172
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 1.1184573002754821e-05,
+      "loss": 0.8327,
+      "step": 173
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.112947658402204e-05,
+      "loss": 0.8157,
+      "step": 174
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1.1074380165289258e-05,
+      "loss": 0.834,
+      "step": 175
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1.1019283746556476e-05,
+      "loss": 0.8091,
+      "step": 176
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.0964187327823692e-05,
+      "loss": 0.8677,
+      "step": 177
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.0909090909090909e-05,
+      "loss": 0.8419,
+      "step": 178
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1.0853994490358129e-05,
+      "loss": 0.8508,
+      "step": 179
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 1.0798898071625345e-05,
+      "loss": 0.8057,
+      "step": 180
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.0743801652892562e-05,
+      "loss": 0.8725,
+      "step": 181
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.068870523415978e-05,
+      "loss": 0.8944,
+      "step": 182
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.0633608815426998e-05,
+      "loss": 0.799,
+      "step": 183
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1.0578512396694216e-05,
+      "loss": 0.8201,
+      "step": 184
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1.0523415977961433e-05,
+      "loss": 0.847,
+      "step": 185
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.046831955922865e-05,
+      "loss": 0.8634,
+      "step": 186
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 1.041322314049587e-05,
+      "loss": 0.8602,
+      "step": 187
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 1.0358126721763086e-05,
+      "loss": 0.8378,
+      "step": 188
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.0303030303030304e-05,
+      "loss": 0.8241,
+      "step": 189
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.024793388429752e-05,
+      "loss": 0.8793,
+      "step": 190
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 1.019283746556474e-05,
+      "loss": 0.934,
+      "step": 191
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.0137741046831957e-05,
+      "loss": 0.798,
+      "step": 192
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.0082644628099174e-05,
+      "loss": 0.7906,
+      "step": 193
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 1.0027548209366392e-05,
+      "loss": 0.8968,
+      "step": 194
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 9.97245179063361e-06,
+      "loss": 0.9144,
+      "step": 195
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 9.917355371900828e-06,
+      "loss": 0.7989,
+      "step": 196
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 9.862258953168045e-06,
+      "loss": 0.8241,
+      "step": 197
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 9.807162534435263e-06,
+      "loss": 0.8218,
+      "step": 198
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 9.75206611570248e-06,
+      "loss": 0.8479,
+      "step": 199
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 9.696969696969698e-06,
+      "loss": 0.8388,
+      "step": 200
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 9.641873278236916e-06,
+      "loss": 0.88,
+      "step": 201
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 9.586776859504134e-06,
+      "loss": 0.8438,
+      "step": 202
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 9.53168044077135e-06,
+      "loss": 0.8863,
+      "step": 203
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 9.476584022038569e-06,
+      "loss": 0.8427,
+      "step": 204
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 9.421487603305785e-06,
+      "loss": 0.9552,
+      "step": 205
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 9.366391184573003e-06,
+      "loss": 0.7866,
+      "step": 206
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 9.311294765840222e-06,
+      "loss": 0.8672,
+      "step": 207
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 9.25619834710744e-06,
+      "loss": 0.8376,
+      "step": 208
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 9.201101928374656e-06,
+      "loss": 0.8611,
+      "step": 209
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 9.146005509641875e-06,
+      "loss": 0.8543,
+      "step": 210
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.7672,
+      "step": 211
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 9.03581267217631e-06,
+      "loss": 0.8341,
+      "step": 212
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 8.980716253443526e-06,
+      "loss": 0.8354,
+      "step": 213
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 8.925619834710744e-06,
+      "loss": 0.8699,
+      "step": 214
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 8.870523415977962e-06,
+      "loss": 0.8848,
+      "step": 215
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 8.81542699724518e-06,
+      "loss": 0.8109,
+      "step": 216
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 8.760330578512397e-06,
+      "loss": 0.8197,
+      "step": 217
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 8.705234159779615e-06,
+      "loss": 0.8186,
+      "step": 218
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 8.650137741046832e-06,
+      "loss": 0.8382,
+      "step": 219
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 8.59504132231405e-06,
+      "loss": 0.8341,
+      "step": 220
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 8.539944903581268e-06,
+      "loss": 0.8151,
+      "step": 221
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 8.484848484848486e-06,
+      "loss": 0.8597,
+      "step": 222
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 8.429752066115703e-06,
+      "loss": 0.8468,
+      "step": 223
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 8.374655647382921e-06,
+      "loss": 0.8361,
+      "step": 224
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 8.319559228650137e-06,
+      "loss": 0.8236,
+      "step": 225
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 8.264462809917356e-06,
+      "loss": 0.824,
+      "step": 226
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 8.209366391184574e-06,
+      "loss": 0.8553,
+      "step": 227
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 8.154269972451792e-06,
+      "loss": 0.9107,
+      "step": 228
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 8.099173553719009e-06,
+      "loss": 0.828,
+      "step": 229
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 8.044077134986227e-06,
+      "loss": 0.8415,
+      "step": 230
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 7.988980716253443e-06,
+      "loss": 0.8187,
+      "step": 231
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 7.933884297520661e-06,
+      "loss": 0.8463,
+      "step": 232
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 7.87878787878788e-06,
+      "loss": 0.8565,
+      "step": 233
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 7.823691460055098e-06,
+      "loss": 0.8973,
+      "step": 234
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 7.768595041322314e-06,
+      "loss": 0.8126,
+      "step": 235
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 7.713498622589533e-06,
+      "loss": 0.8808,
+      "step": 236
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 7.658402203856749e-06,
+      "loss": 0.8257,
+      "step": 237
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 7.603305785123968e-06,
+      "loss": 0.7625,
+      "step": 238
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 7.548209366391185e-06,
+      "loss": 0.787,
+      "step": 239
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 7.493112947658403e-06,
+      "loss": 0.8942,
+      "step": 240
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 7.43801652892562e-06,
+      "loss": 0.8373,
+      "step": 241
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 7.3829201101928385e-06,
+      "loss": 0.8113,
+      "step": 242
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 7.327823691460055e-06,
+      "loss": 0.8187,
+      "step": 243
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 7.272727272727273e-06,
+      "loss": 0.8349,
+      "step": 244
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 7.2176308539944905e-06,
+      "loss": 0.8702,
+      "step": 245
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 7.162534435261709e-06,
+      "loss": 0.8888,
+      "step": 246
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 7.107438016528926e-06,
+      "loss": 0.8853,
+      "step": 247
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 7.052341597796144e-06,
+      "loss": 0.8767,
+      "step": 248
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 6.997245179063361e-06,
+      "loss": 0.8292,
+      "step": 249
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 6.942148760330579e-06,
+      "loss": 0.8609,
+      "step": 250
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 6.887052341597796e-06,
+      "loss": 0.6381,
+      "step": 251
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 6.8319559228650146e-06,
+      "loss": 0.6476,
+      "step": 252
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 6.776859504132232e-06,
+      "loss": 0.623,
+      "step": 253
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 6.72176308539945e-06,
+      "loss": 0.5721,
+      "step": 254
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.6855,
+      "step": 255
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 6.611570247933885e-06,
+      "loss": 0.6758,
+      "step": 256
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 6.556473829201102e-06,
+      "loss": 0.686,
+      "step": 257
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 6.50137741046832e-06,
+      "loss": 0.605,
+      "step": 258
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 6.446280991735537e-06,
+      "loss": 0.6356,
+      "step": 259
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 6.391184573002755e-06,
+      "loss": 0.6412,
+      "step": 260
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 6.3360881542699725e-06,
+      "loss": 0.5821,
+      "step": 261
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 6.280991735537191e-06,
+      "loss": 0.6579,
+      "step": 262
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 6.225895316804408e-06,
+      "loss": 0.6324,
+      "step": 263
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 6.170798898071626e-06,
+      "loss": 0.6206,
+      "step": 264
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 6.115702479338843e-06,
+      "loss": 0.568,
+      "step": 265
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 6.060606060606061e-06,
+      "loss": 0.6497,
+      "step": 266
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 6.005509641873278e-06,
+      "loss": 0.7128,
+      "step": 267
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 5.9504132231404965e-06,
+      "loss": 0.6326,
+      "step": 268
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 5.895316804407714e-06,
+      "loss": 0.6711,
+      "step": 269
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 5.840220385674932e-06,
+      "loss": 0.6052,
+      "step": 270
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 5.785123966942149e-06,
+      "loss": 0.616,
+      "step": 271
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 5.730027548209367e-06,
+      "loss": 0.6522,
+      "step": 272
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 5.674931129476584e-06,
+      "loss": 0.6183,
+      "step": 273
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 5.619834710743802e-06,
+      "loss": 0.6063,
+      "step": 274
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 5.56473829201102e-06,
+      "loss": 0.6638,
+      "step": 275
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 5.509641873278238e-06,
+      "loss": 0.6629,
+      "step": 276
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 5.4545454545454545e-06,
+      "loss": 0.6219,
+      "step": 277
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 5.399449035812673e-06,
+      "loss": 0.6322,
+      "step": 278
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 5.34435261707989e-06,
+      "loss": 0.6001,
+      "step": 279
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 5.289256198347108e-06,
+      "loss": 0.6135,
+      "step": 280
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 5.234159779614325e-06,
+      "loss": 0.6853,
+      "step": 281
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 5.179063360881543e-06,
+      "loss": 0.6594,
+      "step": 282
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 5.12396694214876e-06,
+      "loss": 0.5779,
+      "step": 283
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 5.0688705234159785e-06,
+      "loss": 0.6122,
+      "step": 284
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 5.013774104683196e-06,
+      "loss": 0.5967,
+      "step": 285
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 4.958677685950414e-06,
+      "loss": 0.634,
+      "step": 286
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 4.903581267217631e-06,
+      "loss": 0.5878,
+      "step": 287
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 4.848484848484849e-06,
+      "loss": 0.6064,
+      "step": 288
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 4.793388429752067e-06,
+      "loss": 0.6401,
+      "step": 289
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 4.738292011019284e-06,
+      "loss": 0.6212,
+      "step": 290
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 4.683195592286502e-06,
+      "loss": 0.675,
+      "step": 291
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 4.62809917355372e-06,
+      "loss": 0.5758,
+      "step": 292
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 4.573002754820937e-06,
+      "loss": 0.6208,
+      "step": 293
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 4.517906336088155e-06,
+      "loss": 0.6623,
+      "step": 294
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 4.462809917355372e-06,
+      "loss": 0.7006,
+      "step": 295
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 4.40771349862259e-06,
+      "loss": 0.6446,
+      "step": 296
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 4.3526170798898075e-06,
+      "loss": 0.595,
+      "step": 297
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 4.297520661157025e-06,
+      "loss": 0.6179,
+      "step": 298
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 4.242424242424243e-06,
+      "loss": 0.6056,
+      "step": 299
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 4.1873278236914605e-06,
+      "loss": 0.6975,
+      "step": 300
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 4.132231404958678e-06,
+      "loss": 0.577,
+      "step": 301
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 4.077134986225896e-06,
+      "loss": 0.6164,
+      "step": 302
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 4.022038567493113e-06,
+      "loss": 0.5844,
+      "step": 303
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 3.966942148760331e-06,
+      "loss": 0.6092,
+      "step": 304
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 3.911845730027549e-06,
+      "loss": 0.6448,
+      "step": 305
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 3.856749311294766e-06,
+      "loss": 0.6078,
+      "step": 306
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 3.801652892561984e-06,
+      "loss": 0.67,
+      "step": 307
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 3.7465564738292014e-06,
+      "loss": 0.6513,
+      "step": 308
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 3.6914600550964192e-06,
+      "loss": 0.6909,
+      "step": 309
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 3.6363636363636366e-06,
+      "loss": 0.6293,
+      "step": 310
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 3.5812672176308544e-06,
+      "loss": 0.5398,
+      "step": 311
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 3.526170798898072e-06,
+      "loss": 0.6252,
+      "step": 312
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 3.4710743801652895e-06,
+      "loss": 0.5939,
+      "step": 313
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 3.4159779614325073e-06,
+      "loss": 0.628,
+      "step": 314
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 3.360881542699725e-06,
+      "loss": 0.5991,
+      "step": 315
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 3.3057851239669424e-06,
+      "loss": 0.6168,
+      "step": 316
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 3.25068870523416e-06,
+      "loss": 0.6345,
+      "step": 317
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 3.1955922865013776e-06,
+      "loss": 0.6272,
+      "step": 318
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 3.1404958677685953e-06,
+      "loss": 0.6319,
+      "step": 319
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 3.085399449035813e-06,
+      "loss": 0.6536,
+      "step": 320
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 3.0303030303030305e-06,
+      "loss": 0.5719,
+      "step": 321
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 2.9752066115702483e-06,
+      "loss": 0.5611,
+      "step": 322
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 2.920110192837466e-06,
+      "loss": 0.6402,
+      "step": 323
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 2.8650137741046834e-06,
+      "loss": 0.6184,
+      "step": 324
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 2.809917355371901e-06,
+      "loss": 0.6345,
+      "step": 325
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 2.754820936639119e-06,
+      "loss": 0.6637,
+      "step": 326
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 2.6997245179063363e-06,
+      "loss": 0.5998,
+      "step": 327
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 2.644628099173554e-06,
+      "loss": 0.5679,
+      "step": 328
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 2.5895316804407715e-06,
+      "loss": 0.6406,
+      "step": 329
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 2.5344352617079892e-06,
+      "loss": 0.633,
+      "step": 330
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 2.479338842975207e-06,
+      "loss": 0.6439,
+      "step": 331
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 2.4242424242424244e-06,
+      "loss": 0.6216,
+      "step": 332
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 2.369146005509642e-06,
+      "loss": 0.6505,
+      "step": 333
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 2.31404958677686e-06,
+      "loss": 0.6455,
+      "step": 334
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 2.2589531680440773e-06,
+      "loss": 0.6273,
+      "step": 335
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 2.203856749311295e-06,
+      "loss": 0.6205,
+      "step": 336
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 2.1487603305785124e-06,
+      "loss": 0.5994,
+      "step": 337
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 2.0936639118457302e-06,
+      "loss": 0.5779,
+      "step": 338
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 2.038567493112948e-06,
+      "loss": 0.6437,
+      "step": 339
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 1.9834710743801654e-06,
+      "loss": 0.6416,
+      "step": 340
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 1.928374655647383e-06,
+      "loss": 0.6012,
+      "step": 341
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 1.8732782369146007e-06,
+      "loss": 0.6741,
+      "step": 342
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 1.8181818181818183e-06,
+      "loss": 0.6456,
+      "step": 343
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 1.763085399449036e-06,
+      "loss": 0.6504,
+      "step": 344
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 1.7079889807162536e-06,
+      "loss": 0.6183,
+      "step": 345
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 1.6528925619834712e-06,
+      "loss": 0.6371,
+      "step": 346
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 1.5977961432506888e-06,
+      "loss": 0.665,
+      "step": 347
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 1.5426997245179066e-06,
+      "loss": 0.6632,
+      "step": 348
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 1.4876033057851241e-06,
+      "loss": 0.6072,
+      "step": 349
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 1.4325068870523417e-06,
+      "loss": 0.6207,
+      "step": 350
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 1.3774104683195595e-06,
+      "loss": 0.5713,
+      "step": 351
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 1.322314049586777e-06,
+      "loss": 0.6021,
+      "step": 352
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 1.2672176308539946e-06,
+      "loss": 0.6223,
+      "step": 353
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 1.2121212121212122e-06,
+      "loss": 0.6316,
+      "step": 354
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 1.15702479338843e-06,
+      "loss": 0.6586,
+      "step": 355
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 1.1019283746556475e-06,
+      "loss": 0.6332,
+      "step": 356
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 1.0468319559228651e-06,
+      "loss": 0.682,
+      "step": 357
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 9.917355371900827e-07,
+      "loss": 0.6285,
+      "step": 358
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 9.366391184573004e-07,
+      "loss": 0.616,
+      "step": 359
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 8.81542699724518e-07,
+      "loss": 0.6497,
+      "step": 360
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 8.264462809917356e-07,
+      "loss": 0.6013,
+      "step": 361
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 7.713498622589533e-07,
+      "loss": 0.6165,
+      "step": 362
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 7.162534435261709e-07,
+      "loss": 0.6757,
+      "step": 363
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 6.611570247933885e-07,
+      "loss": 0.6242,
+      "step": 364
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 6.060606060606061e-07,
+      "loss": 0.6069,
+      "step": 365
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 5.509641873278238e-07,
+      "loss": 0.629,
+      "step": 366
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 4.958677685950413e-07,
+      "loss": 0.6581,
+      "step": 367
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 4.40771349862259e-07,
+      "loss": 0.5775,
+      "step": 368
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 3.8567493112947664e-07,
+      "loss": 0.6631,
+      "step": 369
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 3.3057851239669426e-07,
+      "loss": 0.5838,
+      "step": 370
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 2.754820936639119e-07,
+      "loss": 0.63,
+      "step": 371
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 2.203856749311295e-07,
+      "loss": 0.6276,
+      "step": 372
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 1.6528925619834713e-07,
+      "loss": 0.6565,
+      "step": 373
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 1.1019283746556475e-07,
+      "loss": 0.6356,
+      "step": 374
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 5.509641873278238e-08,
+      "loss": 0.6329,
+      "step": 375
+    },
+    {
+      "epoch": 3.0,
+      "step": 375,
+      "total_flos": 28638626365440.0,
+      "train_loss": 0.8634774754842123,
+      "train_runtime": 6551.4446,
+      "train_samples_per_second": 7.327,
+      "train_steps_per_second": 0.057
+    }
+  ],
+  "max_steps": 375,
+  "num_train_epochs": 3,
+  "total_flos": 28638626365440.0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:794e4c78cfb3077d1c3a5e8aad802d304d1970b083809c1a39083a3bdd3dacff
+size 5051

zero_to_fp32.py

@@ -0,0 +1,483 @@
+#!/usr/bin/env python
+'''Copyright The Microsoft DeepSpeed Team'''
+
+# 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)