Upload model (#2)

- Upload model (24b64bdbdae58cb33d2e94452239671dbe8f3af4)


Co-authored-by: Romrawin Chumpu <romrawinjp@users.noreply.huggingface.co>

added_tokens.json

@@ -0,0 +1,7 @@
+{
+  "<CLS>": 41070,
+  "<EOD>": 41072,
+  "<MASK>": 41073,
+  "<PAD>": 41074,
+  "<SEP>": 41071
+}

config.json

@@ -0,0 +1,28 @@
+{
+  "_name_or_path": "/project/lt900048-ai24tn/models/openthaigpt/openthaigpt-1.0.0-7b-chat",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.40.1",
+  "use_cache": false,
+  "vocab_size": 41088
+}

latest

@@ -0,0 +1 @@
+global_step285

special_tokens_map.json

@@ -0,0 +1,29 @@
+{
+  "additional_special_tokens": [
+    "<unk>",
+    "<s>",
+    "</s>"
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "</s>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02df43dcae8c7b5b122d45f642e42c96577cdd09fd949c6996051886c72ab002
+size 717508

tokenizer_config.json

@@ -0,0 +1,88 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "41070": {
+      "content": "<CLS>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "41071": {
+      "content": "<SEP>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "41072": {
+      "content": "<EOD>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "41073": {
+      "content": "<MASK>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "41074": {
+      "content": "<PAD>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<unk>",
+    "<s>",
+    "</s>"
+  ],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 2048,
+  "pad_token": "</s>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": true
+}

trainer_state.json

@@ -0,0 +1,2025 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 285,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0035087719298245615,
+      "grad_norm": 18.65219621501764,
+      "learning_rate": 0.0,
+      "loss": 1.9653,
+      "step": 1
+    },
+    {
+      "epoch": 0.007017543859649123,
+      "grad_norm": 5.5068522642682325,
+      "learning_rate": 2.5237190142858296e-05,
+      "loss": 1.4912,
+      "step": 2
+    },
+    {
+      "epoch": 0.010526315789473684,
+      "grad_norm": 5.300187955783763,
+      "learning_rate": 4e-05,
+      "loss": 1.488,
+      "step": 3
+    },
+    {
+      "epoch": 0.014035087719298246,
+      "grad_norm": 3.607547540937298,
+      "learning_rate": 5.047438028571659e-05,
+      "loss": 1.3376,
+      "step": 4
+    },
+    {
+      "epoch": 0.017543859649122806,
+      "grad_norm": 2.8856871756952254,
+      "learning_rate": 5.859894082871708e-05,
+      "loss": 1.2775,
+      "step": 5
+    },
+    {
+      "epoch": 0.021052631578947368,
+      "grad_norm": 2.1943773684803283,
+      "learning_rate": 6.52371901428583e-05,
+      "loss": 1.1621,
+      "step": 6
+    },
+    {
+      "epoch": 0.02456140350877193,
+      "grad_norm": 1.6148482128986932,
+      "learning_rate": 7.08497499664569e-05,
+      "loss": 1.1393,
+      "step": 7
+    },
+    {
+      "epoch": 0.028070175438596492,
+      "grad_norm": 1.4378600160323183,
+      "learning_rate": 7.571157042857488e-05,
+      "loss": 1.1342,
+      "step": 8
+    },
+    {
+      "epoch": 0.031578947368421054,
+      "grad_norm": 1.3979572631886956,
+      "learning_rate": 8e-05,
+      "loss": 1.1196,
+      "step": 9
+    },
+    {
+      "epoch": 0.03508771929824561,
+      "grad_norm": 1.3666278253029815,
+      "learning_rate": 8e-05,
+      "loss": 1.1479,
+      "step": 10
+    },
+    {
+      "epoch": 0.03859649122807018,
+      "grad_norm": 1.2483549253543214,
+      "learning_rate": 7.971014492753623e-05,
+      "loss": 1.1125,
+      "step": 11
+    },
+    {
+      "epoch": 0.042105263157894736,
+      "grad_norm": 1.347256910957925,
+      "learning_rate": 7.942028985507246e-05,
+      "loss": 1.1588,
+      "step": 12
+    },
+    {
+      "epoch": 0.0456140350877193,
+      "grad_norm": 1.3748876632936469,
+      "learning_rate": 7.91304347826087e-05,
+      "loss": 1.0846,
+      "step": 13
+    },
+    {
+      "epoch": 0.04912280701754386,
+      "grad_norm": 1.329787006205099,
+      "learning_rate": 7.884057971014494e-05,
+      "loss": 1.1839,
+      "step": 14
+    },
+    {
+      "epoch": 0.05263157894736842,
+      "grad_norm": 1.1855019705035863,
+      "learning_rate": 7.855072463768117e-05,
+      "loss": 1.1546,
+      "step": 15
+    },
+    {
+      "epoch": 0.056140350877192984,
+      "grad_norm": 1.2020228064951217,
+      "learning_rate": 7.82608695652174e-05,
+      "loss": 1.0999,
+      "step": 16
+    },
+    {
+      "epoch": 0.05964912280701754,
+      "grad_norm": 1.2879067395522685,
+      "learning_rate": 7.797101449275363e-05,
+      "loss": 1.1466,
+      "step": 17
+    },
+    {
+      "epoch": 0.06315789473684211,
+      "grad_norm": 1.256590992120541,
+      "learning_rate": 7.768115942028987e-05,
+      "loss": 1.1593,
+      "step": 18
+    },
+    {
+      "epoch": 0.06666666666666667,
+      "grad_norm": 1.1654027316591642,
+      "learning_rate": 7.73913043478261e-05,
+      "loss": 1.1806,
+      "step": 19
+    },
+    {
+      "epoch": 0.07017543859649122,
+      "grad_norm": 1.2007307897010868,
+      "learning_rate": 7.710144927536232e-05,
+      "loss": 1.1416,
+      "step": 20
+    },
+    {
+      "epoch": 0.07368421052631578,
+      "grad_norm": 1.1499545393774593,
+      "learning_rate": 7.681159420289856e-05,
+      "loss": 1.1203,
+      "step": 21
+    },
+    {
+      "epoch": 0.07719298245614035,
+      "grad_norm": 1.0810536737354983,
+      "learning_rate": 7.652173913043479e-05,
+      "loss": 1.0875,
+      "step": 22
+    },
+    {
+      "epoch": 0.08070175438596491,
+      "grad_norm": 1.249217128847186,
+      "learning_rate": 7.623188405797102e-05,
+      "loss": 1.1878,
+      "step": 23
+    },
+    {
+      "epoch": 0.08421052631578947,
+      "grad_norm": 1.171401194067982,
+      "learning_rate": 7.594202898550726e-05,
+      "loss": 1.1416,
+      "step": 24
+    },
+    {
+      "epoch": 0.08771929824561403,
+      "grad_norm": 1.1864422136018933,
+      "learning_rate": 7.565217391304349e-05,
+      "loss": 1.1559,
+      "step": 25
+    },
+    {
+      "epoch": 0.0912280701754386,
+      "grad_norm": 1.1084100899911409,
+      "learning_rate": 7.536231884057971e-05,
+      "loss": 1.127,
+      "step": 26
+    },
+    {
+      "epoch": 0.09473684210526316,
+      "grad_norm": 1.0848431056969332,
+      "learning_rate": 7.507246376811594e-05,
+      "loss": 1.1621,
+      "step": 27
+    },
+    {
+      "epoch": 0.09824561403508772,
+      "grad_norm": 1.160465065160364,
+      "learning_rate": 7.478260869565218e-05,
+      "loss": 1.1086,
+      "step": 28
+    },
+    {
+      "epoch": 0.10175438596491228,
+      "grad_norm": 1.0888824123502896,
+      "learning_rate": 7.449275362318841e-05,
+      "loss": 1.1402,
+      "step": 29
+    },
+    {
+      "epoch": 0.10526315789473684,
+      "grad_norm": 1.136530946229479,
+      "learning_rate": 7.420289855072465e-05,
+      "loss": 1.1533,
+      "step": 30
+    },
+    {
+      "epoch": 0.10877192982456141,
+      "grad_norm": 1.1193975180437097,
+      "learning_rate": 7.391304347826088e-05,
+      "loss": 1.1194,
+      "step": 31
+    },
+    {
+      "epoch": 0.11228070175438597,
+      "grad_norm": 1.032391835314362,
+      "learning_rate": 7.36231884057971e-05,
+      "loss": 1.086,
+      "step": 32
+    },
+    {
+      "epoch": 0.11578947368421053,
+      "grad_norm": 1.1734719012857107,
+      "learning_rate": 7.333333333333333e-05,
+      "loss": 1.1515,
+      "step": 33
+    },
+    {
+      "epoch": 0.11929824561403508,
+      "grad_norm": 1.1223429839275594,
+      "learning_rate": 7.304347826086957e-05,
+      "loss": 1.1067,
+      "step": 34
+    },
+    {
+      "epoch": 0.12280701754385964,
+      "grad_norm": 1.0919749212433656,
+      "learning_rate": 7.27536231884058e-05,
+      "loss": 1.1298,
+      "step": 35
+    },
+    {
+      "epoch": 0.12631578947368421,
+      "grad_norm": 1.061313889855867,
+      "learning_rate": 7.246376811594204e-05,
+      "loss": 1.1323,
+      "step": 36
+    },
+    {
+      "epoch": 0.12982456140350876,
+      "grad_norm": 0.9886502393238654,
+      "learning_rate": 7.217391304347827e-05,
+      "loss": 1.1358,
+      "step": 37
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 1.090149070089068,
+      "learning_rate": 7.18840579710145e-05,
+      "loss": 1.1083,
+      "step": 38
+    },
+    {
+      "epoch": 0.1368421052631579,
+      "grad_norm": 1.2141824886927202,
+      "learning_rate": 7.159420289855072e-05,
+      "loss": 1.1659,
+      "step": 39
+    },
+    {
+      "epoch": 0.14035087719298245,
+      "grad_norm": 1.2804981481033841,
+      "learning_rate": 7.130434782608696e-05,
+      "loss": 1.1769,
+      "step": 40
+    },
+    {
+      "epoch": 0.14385964912280702,
+      "grad_norm": 1.085391641580103,
+      "learning_rate": 7.101449275362319e-05,
+      "loss": 1.1307,
+      "step": 41
+    },
+    {
+      "epoch": 0.14736842105263157,
+      "grad_norm": 1.0432511632089205,
+      "learning_rate": 7.072463768115943e-05,
+      "loss": 1.1152,
+      "step": 42
+    },
+    {
+      "epoch": 0.15087719298245614,
+      "grad_norm": 1.0805636921488617,
+      "learning_rate": 7.043478260869566e-05,
+      "loss": 1.1131,
+      "step": 43
+    },
+    {
+      "epoch": 0.1543859649122807,
+      "grad_norm": 1.0831222641710494,
+      "learning_rate": 7.014492753623189e-05,
+      "loss": 1.1585,
+      "step": 44
+    },
+    {
+      "epoch": 0.15789473684210525,
+      "grad_norm": 0.9919690920164701,
+      "learning_rate": 6.985507246376811e-05,
+      "loss": 1.1235,
+      "step": 45
+    },
+    {
+      "epoch": 0.16140350877192983,
+      "grad_norm": 0.9498489248223517,
+      "learning_rate": 6.956521739130436e-05,
+      "loss": 1.0856,
+      "step": 46
+    },
+    {
+      "epoch": 0.1649122807017544,
+      "grad_norm": 1.0276715714307674,
+      "learning_rate": 6.927536231884058e-05,
+      "loss": 1.1378,
+      "step": 47
+    },
+    {
+      "epoch": 0.16842105263157894,
+      "grad_norm": 1.094458012231917,
+      "learning_rate": 6.898550724637682e-05,
+      "loss": 1.117,
+      "step": 48
+    },
+    {
+      "epoch": 0.17192982456140352,
+      "grad_norm": 1.089782170199694,
+      "learning_rate": 6.869565217391305e-05,
+      "loss": 1.1471,
+      "step": 49
+    },
+    {
+      "epoch": 0.17543859649122806,
+      "grad_norm": 1.0908602591310133,
+      "learning_rate": 6.840579710144928e-05,
+      "loss": 1.1711,
+      "step": 50
+    },
+    {
+      "epoch": 0.17894736842105263,
+      "grad_norm": 1.126099914721245,
+      "learning_rate": 6.811594202898552e-05,
+      "loss": 1.1437,
+      "step": 51
+    },
+    {
+      "epoch": 0.1824561403508772,
+      "grad_norm": 1.1004469394026717,
+      "learning_rate": 6.782608695652175e-05,
+      "loss": 1.1502,
+      "step": 52
+    },
+    {
+      "epoch": 0.18596491228070175,
+      "grad_norm": 1.0690976698679064,
+      "learning_rate": 6.753623188405797e-05,
+      "loss": 1.1436,
+      "step": 53
+    },
+    {
+      "epoch": 0.18947368421052632,
+      "grad_norm": 1.0893372270742567,
+      "learning_rate": 6.72463768115942e-05,
+      "loss": 1.1028,
+      "step": 54
+    },
+    {
+      "epoch": 0.19298245614035087,
+      "grad_norm": 1.038406028318434,
+      "learning_rate": 6.695652173913044e-05,
+      "loss": 1.1601,
+      "step": 55
+    },
+    {
+      "epoch": 0.19649122807017544,
+      "grad_norm": 0.9960149874705063,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 1.102,
+      "step": 56
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.073973228867499,
+      "learning_rate": 6.637681159420291e-05,
+      "loss": 1.1213,
+      "step": 57
+    },
+    {
+      "epoch": 0.20350877192982456,
+      "grad_norm": 1.1952354445407958,
+      "learning_rate": 6.608695652173914e-05,
+      "loss": 1.1046,
+      "step": 58
+    },
+    {
+      "epoch": 0.20701754385964913,
+      "grad_norm": 1.0175331854073473,
+      "learning_rate": 6.579710144927537e-05,
+      "loss": 1.1192,
+      "step": 59
+    },
+    {
+      "epoch": 0.21052631578947367,
+      "grad_norm": 1.053567224648277,
+      "learning_rate": 6.550724637681159e-05,
+      "loss": 1.1514,
+      "step": 60
+    },
+    {
+      "epoch": 0.21403508771929824,
+      "grad_norm": 1.1549538449027954,
+      "learning_rate": 6.521739130434783e-05,
+      "loss": 1.161,
+      "step": 61
+    },
+    {
+      "epoch": 0.21754385964912282,
+      "grad_norm": 0.9980361090477836,
+      "learning_rate": 6.492753623188406e-05,
+      "loss": 1.0715,
+      "step": 62
+    },
+    {
+      "epoch": 0.22105263157894736,
+      "grad_norm": 1.0664047714210805,
+      "learning_rate": 6.46376811594203e-05,
+      "loss": 1.172,
+      "step": 63
+    },
+    {
+      "epoch": 0.22456140350877193,
+      "grad_norm": 1.0338684936573976,
+      "learning_rate": 6.434782608695653e-05,
+      "loss": 1.0547,
+      "step": 64
+    },
+    {
+      "epoch": 0.22807017543859648,
+      "grad_norm": 1.0164199027592216,
+      "learning_rate": 6.405797101449276e-05,
+      "loss": 1.0821,
+      "step": 65
+    },
+    {
+      "epoch": 0.23157894736842105,
+      "grad_norm": 1.08287922650492,
+      "learning_rate": 6.376811594202898e-05,
+      "loss": 1.112,
+      "step": 66
+    },
+    {
+      "epoch": 0.23508771929824562,
+      "grad_norm": 1.012260870534961,
+      "learning_rate": 6.347826086956523e-05,
+      "loss": 1.1339,
+      "step": 67
+    },
+    {
+      "epoch": 0.23859649122807017,
+      "grad_norm": 0.9284957156873009,
+      "learning_rate": 6.318840579710145e-05,
+      "loss": 1.0655,
+      "step": 68
+    },
+    {
+      "epoch": 0.24210526315789474,
+      "grad_norm": 0.9699566883456087,
+      "learning_rate": 6.28985507246377e-05,
+      "loss": 1.0447,
+      "step": 69
+    },
+    {
+      "epoch": 0.24561403508771928,
+      "grad_norm": 0.9953128820241032,
+      "learning_rate": 6.260869565217392e-05,
+      "loss": 1.0979,
+      "step": 70
+    },
+    {
+      "epoch": 0.24912280701754386,
+      "grad_norm": 1.2255506094662556,
+      "learning_rate": 6.231884057971015e-05,
+      "loss": 1.1266,
+      "step": 71
+    },
+    {
+      "epoch": 0.25263157894736843,
+      "grad_norm": 1.1053469341884643,
+      "learning_rate": 6.202898550724638e-05,
+      "loss": 1.1083,
+      "step": 72
+    },
+    {
+      "epoch": 0.256140350877193,
+      "grad_norm": 1.0471395283209621,
+      "learning_rate": 6.173913043478262e-05,
+      "loss": 1.1081,
+      "step": 73
+    },
+    {
+      "epoch": 0.2596491228070175,
+      "grad_norm": 1.1372933010569457,
+      "learning_rate": 6.144927536231884e-05,
+      "loss": 1.1191,
+      "step": 74
+    },
+    {
+      "epoch": 0.2631578947368421,
+      "grad_norm": 1.094007360015913,
+      "learning_rate": 6.115942028985507e-05,
+      "loss": 1.1498,
+      "step": 75
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 1.037731837007179,
+      "learning_rate": 6.086956521739131e-05,
+      "loss": 1.078,
+      "step": 76
+    },
+    {
+      "epoch": 0.27017543859649124,
+      "grad_norm": 0.9878690341929469,
+      "learning_rate": 6.057971014492755e-05,
+      "loss": 1.0925,
+      "step": 77
+    },
+    {
+      "epoch": 0.2736842105263158,
+      "grad_norm": 1.0251650017115257,
+      "learning_rate": 6.0289855072463774e-05,
+      "loss": 1.0992,
+      "step": 78
+    },
+    {
+      "epoch": 0.2771929824561403,
+      "grad_norm": 0.9677079627985357,
+      "learning_rate": 6.000000000000001e-05,
+      "loss": 1.0228,
+      "step": 79
+    },
+    {
+      "epoch": 0.2807017543859649,
+      "grad_norm": 1.03084630118753,
+      "learning_rate": 5.9710144927536236e-05,
+      "loss": 1.1026,
+      "step": 80
+    },
+    {
+      "epoch": 0.28421052631578947,
+      "grad_norm": 1.149025536043054,
+      "learning_rate": 5.942028985507246e-05,
+      "loss": 1.0971,
+      "step": 81
+    },
+    {
+      "epoch": 0.28771929824561404,
+      "grad_norm": 0.8977367483124615,
+      "learning_rate": 5.91304347826087e-05,
+      "loss": 1.0665,
+      "step": 82
+    },
+    {
+      "epoch": 0.2912280701754386,
+      "grad_norm": 0.982319885338846,
+      "learning_rate": 5.884057971014494e-05,
+      "loss": 1.0987,
+      "step": 83
+    },
+    {
+      "epoch": 0.29473684210526313,
+      "grad_norm": 0.8863683819755959,
+      "learning_rate": 5.8550724637681166e-05,
+      "loss": 1.0603,
+      "step": 84
+    },
+    {
+      "epoch": 0.2982456140350877,
+      "grad_norm": 1.0566465114428347,
+      "learning_rate": 5.82608695652174e-05,
+      "loss": 1.0961,
+      "step": 85
+    },
+    {
+      "epoch": 0.3017543859649123,
+      "grad_norm": 0.9391730832092359,
+      "learning_rate": 5.797101449275363e-05,
+      "loss": 1.1184,
+      "step": 86
+    },
+    {
+      "epoch": 0.30526315789473685,
+      "grad_norm": 0.9725750087741657,
+      "learning_rate": 5.768115942028986e-05,
+      "loss": 1.1131,
+      "step": 87
+    },
+    {
+      "epoch": 0.3087719298245614,
+      "grad_norm": 0.9134535098625444,
+      "learning_rate": 5.739130434782609e-05,
+      "loss": 1.0635,
+      "step": 88
+    },
+    {
+      "epoch": 0.312280701754386,
+      "grad_norm": 0.9210153237928186,
+      "learning_rate": 5.710144927536232e-05,
+      "loss": 1.0208,
+      "step": 89
+    },
+    {
+      "epoch": 0.3157894736842105,
+      "grad_norm": 1.073076083662152,
+      "learning_rate": 5.681159420289856e-05,
+      "loss": 1.1314,
+      "step": 90
+    },
+    {
+      "epoch": 0.3192982456140351,
+      "grad_norm": 0.9246736723369094,
+      "learning_rate": 5.652173913043479e-05,
+      "loss": 1.0458,
+      "step": 91
+    },
+    {
+      "epoch": 0.32280701754385965,
+      "grad_norm": 0.9415979605613132,
+      "learning_rate": 5.623188405797102e-05,
+      "loss": 1.0432,
+      "step": 92
+    },
+    {
+      "epoch": 0.3263157894736842,
+      "grad_norm": 1.0029090729561863,
+      "learning_rate": 5.594202898550725e-05,
+      "loss": 1.1045,
+      "step": 93
+    },
+    {
+      "epoch": 0.3298245614035088,
+      "grad_norm": 1.0012689311514984,
+      "learning_rate": 5.565217391304348e-05,
+      "loss": 1.0923,
+      "step": 94
+    },
+    {
+      "epoch": 0.3333333333333333,
+      "grad_norm": 0.9253904632856298,
+      "learning_rate": 5.5362318840579714e-05,
+      "loss": 1.028,
+      "step": 95
+    },
+    {
+      "epoch": 0.3368421052631579,
+      "grad_norm": 1.006794875176782,
+      "learning_rate": 5.507246376811594e-05,
+      "loss": 1.0772,
+      "step": 96
+    },
+    {
+      "epoch": 0.34035087719298246,
+      "grad_norm": 0.9652341181001821,
+      "learning_rate": 5.478260869565218e-05,
+      "loss": 1.0544,
+      "step": 97
+    },
+    {
+      "epoch": 0.34385964912280703,
+      "grad_norm": 0.9633053495992878,
+      "learning_rate": 5.449275362318841e-05,
+      "loss": 1.0369,
+      "step": 98
+    },
+    {
+      "epoch": 0.3473684210526316,
+      "grad_norm": 0.889193134564978,
+      "learning_rate": 5.4202898550724644e-05,
+      "loss": 1.0571,
+      "step": 99
+    },
+    {
+      "epoch": 0.3508771929824561,
+      "grad_norm": 1.0688458576983824,
+      "learning_rate": 5.391304347826087e-05,
+      "loss": 1.0875,
+      "step": 100
+    },
+    {
+      "epoch": 0.3543859649122807,
+      "grad_norm": 0.9618220162761959,
+      "learning_rate": 5.3623188405797106e-05,
+      "loss": 1.0989,
+      "step": 101
+    },
+    {
+      "epoch": 0.35789473684210527,
+      "grad_norm": 1.0116504967458333,
+      "learning_rate": 5.333333333333333e-05,
+      "loss": 1.1362,
+      "step": 102
+    },
+    {
+      "epoch": 0.36140350877192984,
+      "grad_norm": 0.9446835004736188,
+      "learning_rate": 5.304347826086957e-05,
+      "loss": 1.0549,
+      "step": 103
+    },
+    {
+      "epoch": 0.3649122807017544,
+      "grad_norm": 1.2634297228252125,
+      "learning_rate": 5.27536231884058e-05,
+      "loss": 1.0865,
+      "step": 104
+    },
+    {
+      "epoch": 0.3684210526315789,
+      "grad_norm": 1.0070752681998931,
+      "learning_rate": 5.2463768115942036e-05,
+      "loss": 1.071,
+      "step": 105
+    },
+    {
+      "epoch": 0.3719298245614035,
+      "grad_norm": 1.0889370997100492,
+      "learning_rate": 5.217391304347826e-05,
+      "loss": 1.1063,
+      "step": 106
+    },
+    {
+      "epoch": 0.37543859649122807,
+      "grad_norm": 0.9667935705662773,
+      "learning_rate": 5.18840579710145e-05,
+      "loss": 1.0666,
+      "step": 107
+    },
+    {
+      "epoch": 0.37894736842105264,
+      "grad_norm": 0.9817859943047047,
+      "learning_rate": 5.1594202898550725e-05,
+      "loss": 1.1025,
+      "step": 108
+    },
+    {
+      "epoch": 0.3824561403508772,
+      "grad_norm": 0.8710066297645505,
+      "learning_rate": 5.130434782608696e-05,
+      "loss": 1.1057,
+      "step": 109
+    },
+    {
+      "epoch": 0.38596491228070173,
+      "grad_norm": 1.0647772098482442,
+      "learning_rate": 5.1014492753623186e-05,
+      "loss": 1.1315,
+      "step": 110
+    },
+    {
+      "epoch": 0.3894736842105263,
+      "grad_norm": 0.9399046787831713,
+      "learning_rate": 5.072463768115943e-05,
+      "loss": 1.0572,
+      "step": 111
+    },
+    {
+      "epoch": 0.3929824561403509,
+      "grad_norm": 0.8768124884597763,
+      "learning_rate": 5.043478260869566e-05,
+      "loss": 1.0401,
+      "step": 112
+    },
+    {
+      "epoch": 0.39649122807017545,
+      "grad_norm": 1.009086025516714,
+      "learning_rate": 5.014492753623189e-05,
+      "loss": 1.0604,
+      "step": 113
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.9419034237826691,
+      "learning_rate": 4.9855072463768116e-05,
+      "loss": 1.0858,
+      "step": 114
+    },
+    {
+      "epoch": 0.40350877192982454,
+      "grad_norm": 0.9876303358742554,
+      "learning_rate": 4.956521739130435e-05,
+      "loss": 1.083,
+      "step": 115
+    },
+    {
+      "epoch": 0.4070175438596491,
+      "grad_norm": 0.934675922910989,
+      "learning_rate": 4.927536231884058e-05,
+      "loss": 1.0749,
+      "step": 116
+    },
+    {
+      "epoch": 0.4105263157894737,
+      "grad_norm": 0.9813286336072813,
+      "learning_rate": 4.898550724637682e-05,
+      "loss": 1.0721,
+      "step": 117
+    },
+    {
+      "epoch": 0.41403508771929826,
+      "grad_norm": 1.0168504955660564,
+      "learning_rate": 4.869565217391305e-05,
+      "loss": 1.0276,
+      "step": 118
+    },
+    {
+      "epoch": 0.41754385964912283,
+      "grad_norm": 0.9087837180593036,
+      "learning_rate": 4.840579710144928e-05,
+      "loss": 1.0104,
+      "step": 119
+    },
+    {
+      "epoch": 0.42105263157894735,
+      "grad_norm": 0.8834303398491904,
+      "learning_rate": 4.8115942028985514e-05,
+      "loss": 1.032,
+      "step": 120
+    },
+    {
+      "epoch": 0.4245614035087719,
+      "grad_norm": 0.9525280076015293,
+      "learning_rate": 4.782608695652174e-05,
+      "loss": 1.0257,
+      "step": 121
+    },
+    {
+      "epoch": 0.4280701754385965,
+      "grad_norm": 0.9493713737636021,
+      "learning_rate": 4.7536231884057976e-05,
+      "loss": 1.0273,
+      "step": 122
+    },
+    {
+      "epoch": 0.43157894736842106,
+      "grad_norm": 0.9098239721172197,
+      "learning_rate": 4.72463768115942e-05,
+      "loss": 0.981,
+      "step": 123
+    },
+    {
+      "epoch": 0.43508771929824563,
+      "grad_norm": 0.9664557596159875,
+      "learning_rate": 4.6956521739130444e-05,
+      "loss": 1.1167,
+      "step": 124
+    },
+    {
+      "epoch": 0.43859649122807015,
+      "grad_norm": 0.8663800542676098,
+      "learning_rate": 4.666666666666667e-05,
+      "loss": 1.0566,
+      "step": 125
+    },
+    {
+      "epoch": 0.4421052631578947,
+      "grad_norm": 0.8587329676076015,
+      "learning_rate": 4.6376811594202906e-05,
+      "loss": 1.0457,
+      "step": 126
+    },
+    {
+      "epoch": 0.4456140350877193,
+      "grad_norm": 1.0139380452518938,
+      "learning_rate": 4.608695652173913e-05,
+      "loss": 1.0871,
+      "step": 127
+    },
+    {
+      "epoch": 0.44912280701754387,
+      "grad_norm": 0.87218312389074,
+      "learning_rate": 4.579710144927537e-05,
+      "loss": 0.9934,
+      "step": 128
+    },
+    {
+      "epoch": 0.45263157894736844,
+      "grad_norm": 0.945265788938204,
+      "learning_rate": 4.5507246376811595e-05,
+      "loss": 1.0598,
+      "step": 129
+    },
+    {
+      "epoch": 0.45614035087719296,
+      "grad_norm": 1.0447712876512982,
+      "learning_rate": 4.521739130434783e-05,
+      "loss": 1.0208,
+      "step": 130
+    },
+    {
+      "epoch": 0.45964912280701753,
+      "grad_norm": 0.8863649966336605,
+      "learning_rate": 4.492753623188406e-05,
+      "loss": 0.9916,
+      "step": 131
+    },
+    {
+      "epoch": 0.4631578947368421,
+      "grad_norm": 0.9389072020523048,
+      "learning_rate": 4.46376811594203e-05,
+      "loss": 1.039,
+      "step": 132
+    },
+    {
+      "epoch": 0.4666666666666667,
+      "grad_norm": 1.00128465703121,
+      "learning_rate": 4.4347826086956525e-05,
+      "loss": 1.007,
+      "step": 133
+    },
+    {
+      "epoch": 0.47017543859649125,
+      "grad_norm": 0.9218948102233011,
+      "learning_rate": 4.405797101449276e-05,
+      "loss": 1.0489,
+      "step": 134
+    },
+    {
+      "epoch": 0.47368421052631576,
+      "grad_norm": 0.8980509099966125,
+      "learning_rate": 4.3768115942028986e-05,
+      "loss": 1.0539,
+      "step": 135
+    },
+    {
+      "epoch": 0.47719298245614034,
+      "grad_norm": 0.8922781997772711,
+      "learning_rate": 4.347826086956522e-05,
+      "loss": 1.0653,
+      "step": 136
+    },
+    {
+      "epoch": 0.4807017543859649,
+      "grad_norm": 0.9229578265481162,
+      "learning_rate": 4.318840579710145e-05,
+      "loss": 1.0359,
+      "step": 137
+    },
+    {
+      "epoch": 0.4842105263157895,
+      "grad_norm": 0.9864176248898312,
+      "learning_rate": 4.289855072463769e-05,
+      "loss": 1.0711,
+      "step": 138
+    },
+    {
+      "epoch": 0.48771929824561405,
+      "grad_norm": 0.9040178292405336,
+      "learning_rate": 4.2608695652173916e-05,
+      "loss": 1.0615,
+      "step": 139
+    },
+    {
+      "epoch": 0.49122807017543857,
+      "grad_norm": 0.9065042000785789,
+      "learning_rate": 4.231884057971015e-05,
+      "loss": 1.0267,
+      "step": 140
+    },
+    {
+      "epoch": 0.49473684210526314,
+      "grad_norm": 0.9289938532156966,
+      "learning_rate": 4.202898550724638e-05,
+      "loss": 1.0269,
+      "step": 141
+    },
+    {
+      "epoch": 0.4982456140350877,
+      "grad_norm": 0.8703398745310128,
+      "learning_rate": 4.173913043478261e-05,
+      "loss": 0.9952,
+      "step": 142
+    },
+    {
+      "epoch": 0.5017543859649123,
+      "grad_norm": 0.9754555079004449,
+      "learning_rate": 4.144927536231884e-05,
+      "loss": 1.0744,
+      "step": 143
+    },
+    {
+      "epoch": 0.5052631578947369,
+      "grad_norm": 1.0698316819851512,
+      "learning_rate": 4.115942028985507e-05,
+      "loss": 1.0179,
+      "step": 144
+    },
+    {
+      "epoch": 0.5087719298245614,
+      "grad_norm": 0.9380968706540932,
+      "learning_rate": 4.0869565217391314e-05,
+      "loss": 0.9932,
+      "step": 145
+    },
+    {
+      "epoch": 0.512280701754386,
+      "grad_norm": 0.9380079488737321,
+      "learning_rate": 4.057971014492754e-05,
+      "loss": 1.0655,
+      "step": 146
+    },
+    {
+      "epoch": 0.5157894736842106,
+      "grad_norm": 0.9764065323167558,
+      "learning_rate": 4.0289855072463776e-05,
+      "loss": 1.0096,
+      "step": 147
+    },
+    {
+      "epoch": 0.519298245614035,
+      "grad_norm": 0.917590332443622,
+      "learning_rate": 4e-05,
+      "loss": 1.0376,
+      "step": 148
+    },
+    {
+      "epoch": 0.5228070175438596,
+      "grad_norm": 0.9466091457378278,
+      "learning_rate": 3.971014492753623e-05,
+      "loss": 1.0722,
+      "step": 149
+    },
+    {
+      "epoch": 0.5263157894736842,
+      "grad_norm": 0.9998149019870339,
+      "learning_rate": 3.942028985507247e-05,
+      "loss": 1.0747,
+      "step": 150
+    },
+    {
+      "epoch": 0.5298245614035088,
+      "grad_norm": 0.9592124249246545,
+      "learning_rate": 3.91304347826087e-05,
+      "loss": 0.9772,
+      "step": 151
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 0.8702377525213307,
+      "learning_rate": 3.884057971014493e-05,
+      "loss": 1.0416,
+      "step": 152
+    },
+    {
+      "epoch": 0.5368421052631579,
+      "grad_norm": 0.9408998854641563,
+      "learning_rate": 3.855072463768116e-05,
+      "loss": 1.053,
+      "step": 153
+    },
+    {
+      "epoch": 0.5403508771929825,
+      "grad_norm": 0.913519043609596,
+      "learning_rate": 3.8260869565217395e-05,
+      "loss": 1.0653,
+      "step": 154
+    },
+    {
+      "epoch": 0.543859649122807,
+      "grad_norm": 1.0325941411560875,
+      "learning_rate": 3.797101449275363e-05,
+      "loss": 1.0248,
+      "step": 155
+    },
+    {
+      "epoch": 0.5473684210526316,
+      "grad_norm": 1.0088857759548826,
+      "learning_rate": 3.7681159420289856e-05,
+      "loss": 1.0984,
+      "step": 156
+    },
+    {
+      "epoch": 0.5508771929824562,
+      "grad_norm": 0.9428277439596274,
+      "learning_rate": 3.739130434782609e-05,
+      "loss": 1.0176,
+      "step": 157
+    },
+    {
+      "epoch": 0.5543859649122806,
+      "grad_norm": 0.9286332315925182,
+      "learning_rate": 3.7101449275362325e-05,
+      "loss": 1.0141,
+      "step": 158
+    },
+    {
+      "epoch": 0.5578947368421052,
+      "grad_norm": 0.8869336003769103,
+      "learning_rate": 3.681159420289855e-05,
+      "loss": 1.0075,
+      "step": 159
+    },
+    {
+      "epoch": 0.5614035087719298,
+      "grad_norm": 0.8749772917176272,
+      "learning_rate": 3.6521739130434786e-05,
+      "loss": 0.9922,
+      "step": 160
+    },
+    {
+      "epoch": 0.5649122807017544,
+      "grad_norm": 0.9177592855322187,
+      "learning_rate": 3.623188405797102e-05,
+      "loss": 1.047,
+      "step": 161
+    },
+    {
+      "epoch": 0.5684210526315789,
+      "grad_norm": 0.9681450836764032,
+      "learning_rate": 3.594202898550725e-05,
+      "loss": 0.9947,
+      "step": 162
+    },
+    {
+      "epoch": 0.5719298245614035,
+      "grad_norm": 0.8944949579910934,
+      "learning_rate": 3.565217391304348e-05,
+      "loss": 1.0414,
+      "step": 163
+    },
+    {
+      "epoch": 0.5754385964912281,
+      "grad_norm": 0.823418699987978,
+      "learning_rate": 3.5362318840579716e-05,
+      "loss": 0.9991,
+      "step": 164
+    },
+    {
+      "epoch": 0.5789473684210527,
+      "grad_norm": 0.9010716175104698,
+      "learning_rate": 3.5072463768115943e-05,
+      "loss": 0.957,
+      "step": 165
+    },
+    {
+      "epoch": 0.5824561403508772,
+      "grad_norm": 0.8553233244911422,
+      "learning_rate": 3.478260869565218e-05,
+      "loss": 0.9758,
+      "step": 166
+    },
+    {
+      "epoch": 0.5859649122807018,
+      "grad_norm": 0.8955709532913586,
+      "learning_rate": 3.449275362318841e-05,
+      "loss": 1.0323,
+      "step": 167
+    },
+    {
+      "epoch": 0.5894736842105263,
+      "grad_norm": 0.9632249383426715,
+      "learning_rate": 3.420289855072464e-05,
+      "loss": 1.0214,
+      "step": 168
+    },
+    {
+      "epoch": 0.5929824561403508,
+      "grad_norm": 1.0070882268479324,
+      "learning_rate": 3.391304347826087e-05,
+      "loss": 1.0443,
+      "step": 169
+    },
+    {
+      "epoch": 0.5964912280701754,
+      "grad_norm": 0.9089812750457459,
+      "learning_rate": 3.36231884057971e-05,
+      "loss": 0.9986,
+      "step": 170
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.9687903432716775,
+      "learning_rate": 3.3333333333333335e-05,
+      "loss": 1.0469,
+      "step": 171
+    },
+    {
+      "epoch": 0.6035087719298246,
+      "grad_norm": 0.9065456102432492,
+      "learning_rate": 3.304347826086957e-05,
+      "loss": 1.0736,
+      "step": 172
+    },
+    {
+      "epoch": 0.6070175438596491,
+      "grad_norm": 0.8191700226087758,
+      "learning_rate": 3.2753623188405796e-05,
+      "loss": 0.9516,
+      "step": 173
+    },
+    {
+      "epoch": 0.6105263157894737,
+      "grad_norm": 0.910478607503674,
+      "learning_rate": 3.246376811594203e-05,
+      "loss": 1.0594,
+      "step": 174
+    },
+    {
+      "epoch": 0.6140350877192983,
+      "grad_norm": 0.9847223979786262,
+      "learning_rate": 3.2173913043478265e-05,
+      "loss": 1.0592,
+      "step": 175
+    },
+    {
+      "epoch": 0.6175438596491228,
+      "grad_norm": 0.9377862201207361,
+      "learning_rate": 3.188405797101449e-05,
+      "loss": 1.0477,
+      "step": 176
+    },
+    {
+      "epoch": 0.6210526315789474,
+      "grad_norm": 0.9077059344710785,
+      "learning_rate": 3.1594202898550726e-05,
+      "loss": 1.0394,
+      "step": 177
+    },
+    {
+      "epoch": 0.624561403508772,
+      "grad_norm": 1.038075786423051,
+      "learning_rate": 3.130434782608696e-05,
+      "loss": 1.0548,
+      "step": 178
+    },
+    {
+      "epoch": 0.6280701754385964,
+      "grad_norm": 0.9160209790965669,
+      "learning_rate": 3.101449275362319e-05,
+      "loss": 0.9976,
+      "step": 179
+    },
+    {
+      "epoch": 0.631578947368421,
+      "grad_norm": 0.8109415841452218,
+      "learning_rate": 3.072463768115942e-05,
+      "loss": 1.0091,
+      "step": 180
+    },
+    {
+      "epoch": 0.6350877192982456,
+      "grad_norm": 0.8492567440061373,
+      "learning_rate": 3.0434782608695656e-05,
+      "loss": 1.0038,
+      "step": 181
+    },
+    {
+      "epoch": 0.6385964912280702,
+      "grad_norm": 0.9589169451188458,
+      "learning_rate": 3.0144927536231887e-05,
+      "loss": 1.0207,
+      "step": 182
+    },
+    {
+      "epoch": 0.6421052631578947,
+      "grad_norm": 0.8848409951191845,
+      "learning_rate": 2.9855072463768118e-05,
+      "loss": 0.9767,
+      "step": 183
+    },
+    {
+      "epoch": 0.6456140350877193,
+      "grad_norm": 0.8955318088948372,
+      "learning_rate": 2.956521739130435e-05,
+      "loss": 1.0078,
+      "step": 184
+    },
+    {
+      "epoch": 0.6491228070175439,
+      "grad_norm": 0.8900555696345945,
+      "learning_rate": 2.9275362318840583e-05,
+      "loss": 1.058,
+      "step": 185
+    },
+    {
+      "epoch": 0.6526315789473685,
+      "grad_norm": 0.9579354937026371,
+      "learning_rate": 2.8985507246376814e-05,
+      "loss": 1.0052,
+      "step": 186
+    },
+    {
+      "epoch": 0.656140350877193,
+      "grad_norm": 0.8773237300221255,
+      "learning_rate": 2.8695652173913044e-05,
+      "loss": 1.0132,
+      "step": 187
+    },
+    {
+      "epoch": 0.6596491228070176,
+      "grad_norm": 0.9290671958630004,
+      "learning_rate": 2.840579710144928e-05,
+      "loss": 0.9978,
+      "step": 188
+    },
+    {
+      "epoch": 0.6631578947368421,
+      "grad_norm": 0.9098091343168585,
+      "learning_rate": 2.811594202898551e-05,
+      "loss": 0.986,
+      "step": 189
+    },
+    {
+      "epoch": 0.6666666666666666,
+      "grad_norm": 0.9494694332915156,
+      "learning_rate": 2.782608695652174e-05,
+      "loss": 1.0194,
+      "step": 190
+    },
+    {
+      "epoch": 0.6701754385964912,
+      "grad_norm": 0.8801338941113726,
+      "learning_rate": 2.753623188405797e-05,
+      "loss": 0.969,
+      "step": 191
+    },
+    {
+      "epoch": 0.6736842105263158,
+      "grad_norm": 0.9510753232451794,
+      "learning_rate": 2.7246376811594205e-05,
+      "loss": 0.9742,
+      "step": 192
+    },
+    {
+      "epoch": 0.6771929824561403,
+      "grad_norm": 0.9589458183246781,
+      "learning_rate": 2.6956521739130436e-05,
+      "loss": 1.0525,
+      "step": 193
+    },
+    {
+      "epoch": 0.6807017543859649,
+      "grad_norm": 0.879997612951922,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 0.9824,
+      "step": 194
+    },
+    {
+      "epoch": 0.6842105263157895,
+      "grad_norm": 0.8389910470867041,
+      "learning_rate": 2.63768115942029e-05,
+      "loss": 1.0056,
+      "step": 195
+    },
+    {
+      "epoch": 0.6877192982456141,
+      "grad_norm": 0.9163566396967311,
+      "learning_rate": 2.608695652173913e-05,
+      "loss": 0.9976,
+      "step": 196
+    },
+    {
+      "epoch": 0.6912280701754386,
+      "grad_norm": 0.9772320516342279,
+      "learning_rate": 2.5797101449275362e-05,
+      "loss": 1.0338,
+      "step": 197
+    },
+    {
+      "epoch": 0.6947368421052632,
+      "grad_norm": 0.960770516269451,
+      "learning_rate": 2.5507246376811593e-05,
+      "loss": 1.0611,
+      "step": 198
+    },
+    {
+      "epoch": 0.6982456140350877,
+      "grad_norm": 0.8890098864104716,
+      "learning_rate": 2.521739130434783e-05,
+      "loss": 1.0168,
+      "step": 199
+    },
+    {
+      "epoch": 0.7017543859649122,
+      "grad_norm": 0.8658933316106927,
+      "learning_rate": 2.4927536231884058e-05,
+      "loss": 1.011,
+      "step": 200
+    },
+    {
+      "epoch": 0.7052631578947368,
+      "grad_norm": 0.8576704263418472,
+      "learning_rate": 2.463768115942029e-05,
+      "loss": 0.9807,
+      "step": 201
+    },
+    {
+      "epoch": 0.7087719298245614,
+      "grad_norm": 0.9186773959277461,
+      "learning_rate": 2.4347826086956526e-05,
+      "loss": 0.9284,
+      "step": 202
+    },
+    {
+      "epoch": 0.712280701754386,
+      "grad_norm": 0.858587966202896,
+      "learning_rate": 2.4057971014492757e-05,
+      "loss": 0.9795,
+      "step": 203
+    },
+    {
+      "epoch": 0.7157894736842105,
+      "grad_norm": 0.9422658689165894,
+      "learning_rate": 2.3768115942028988e-05,
+      "loss": 0.9764,
+      "step": 204
+    },
+    {
+      "epoch": 0.7192982456140351,
+      "grad_norm": 1.0137944232610083,
+      "learning_rate": 2.3478260869565222e-05,
+      "loss": 1.0627,
+      "step": 205
+    },
+    {
+      "epoch": 0.7228070175438597,
+      "grad_norm": 0.9853442185067721,
+      "learning_rate": 2.3188405797101453e-05,
+      "loss": 1.0289,
+      "step": 206
+    },
+    {
+      "epoch": 0.7263157894736842,
+      "grad_norm": 0.8923047177872453,
+      "learning_rate": 2.2898550724637684e-05,
+      "loss": 1.0172,
+      "step": 207
+    },
+    {
+      "epoch": 0.7298245614035088,
+      "grad_norm": 0.8439161677593868,
+      "learning_rate": 2.2608695652173914e-05,
+      "loss": 1.0484,
+      "step": 208
+    },
+    {
+      "epoch": 0.7333333333333333,
+      "grad_norm": 0.8201410283081917,
+      "learning_rate": 2.231884057971015e-05,
+      "loss": 1.0168,
+      "step": 209
+    },
+    {
+      "epoch": 0.7368421052631579,
+      "grad_norm": 0.8442899557106462,
+      "learning_rate": 2.202898550724638e-05,
+      "loss": 0.9728,
+      "step": 210
+    },
+    {
+      "epoch": 0.7403508771929824,
+      "grad_norm": 0.9098630031184369,
+      "learning_rate": 2.173913043478261e-05,
+      "loss": 1.0304,
+      "step": 211
+    },
+    {
+      "epoch": 0.743859649122807,
+      "grad_norm": 0.8217923688708064,
+      "learning_rate": 2.1449275362318844e-05,
+      "loss": 1.0334,
+      "step": 212
+    },
+    {
+      "epoch": 0.7473684210526316,
+      "grad_norm": 0.8420848196609938,
+      "learning_rate": 2.1159420289855075e-05,
+      "loss": 0.9665,
+      "step": 213
+    },
+    {
+      "epoch": 0.7508771929824561,
+      "grad_norm": 0.8810436312516507,
+      "learning_rate": 2.0869565217391306e-05,
+      "loss": 0.9792,
+      "step": 214
+    },
+    {
+      "epoch": 0.7543859649122807,
+      "grad_norm": 0.9749166994556522,
+      "learning_rate": 2.0579710144927537e-05,
+      "loss": 1.0163,
+      "step": 215
+    },
+    {
+      "epoch": 0.7578947368421053,
+      "grad_norm": 0.9619307294480671,
+      "learning_rate": 2.028985507246377e-05,
+      "loss": 1.0276,
+      "step": 216
+    },
+    {
+      "epoch": 0.7614035087719299,
+      "grad_norm": 0.8475320304538961,
+      "learning_rate": 2e-05,
+      "loss": 1.0175,
+      "step": 217
+    },
+    {
+      "epoch": 0.7649122807017544,
+      "grad_norm": 0.9215056645276478,
+      "learning_rate": 1.9710144927536236e-05,
+      "loss": 1.0349,
+      "step": 218
+    },
+    {
+      "epoch": 0.7684210526315789,
+      "grad_norm": 0.8318566015217065,
+      "learning_rate": 1.9420289855072467e-05,
+      "loss": 0.9736,
+      "step": 219
+    },
+    {
+      "epoch": 0.7719298245614035,
+      "grad_norm": 0.8841348427961992,
+      "learning_rate": 1.9130434782608697e-05,
+      "loss": 1.0028,
+      "step": 220
+    },
+    {
+      "epoch": 0.775438596491228,
+      "grad_norm": 0.8726663458778539,
+      "learning_rate": 1.8840579710144928e-05,
+      "loss": 0.9837,
+      "step": 221
+    },
+    {
+      "epoch": 0.7789473684210526,
+      "grad_norm": 0.8933308297868434,
+      "learning_rate": 1.8550724637681162e-05,
+      "loss": 0.9875,
+      "step": 222
+    },
+    {
+      "epoch": 0.7824561403508772,
+      "grad_norm": 0.8473190274519702,
+      "learning_rate": 1.8260869565217393e-05,
+      "loss": 0.9822,
+      "step": 223
+    },
+    {
+      "epoch": 0.7859649122807018,
+      "grad_norm": 0.8786630261494585,
+      "learning_rate": 1.7971014492753624e-05,
+      "loss": 0.9785,
+      "step": 224
+    },
+    {
+      "epoch": 0.7894736842105263,
+      "grad_norm": 0.8769937018569535,
+      "learning_rate": 1.7681159420289858e-05,
+      "loss": 0.9863,
+      "step": 225
+    },
+    {
+      "epoch": 0.7929824561403509,
+      "grad_norm": 0.9418240902774399,
+      "learning_rate": 1.739130434782609e-05,
+      "loss": 0.997,
+      "step": 226
+    },
+    {
+      "epoch": 0.7964912280701755,
+      "grad_norm": 0.9360134126214504,
+      "learning_rate": 1.710144927536232e-05,
+      "loss": 0.9863,
+      "step": 227
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.999458819703,
+      "learning_rate": 1.681159420289855e-05,
+      "loss": 0.9498,
+      "step": 228
+    },
+    {
+      "epoch": 0.8035087719298246,
+      "grad_norm": 0.8238131544325801,
+      "learning_rate": 1.6521739130434785e-05,
+      "loss": 0.9609,
+      "step": 229
+    },
+    {
+      "epoch": 0.8070175438596491,
+      "grad_norm": 0.9110168722541372,
+      "learning_rate": 1.6231884057971015e-05,
+      "loss": 0.9726,
+      "step": 230
+    },
+    {
+      "epoch": 0.8105263157894737,
+      "grad_norm": 0.915280887700176,
+      "learning_rate": 1.5942028985507246e-05,
+      "loss": 0.9687,
+      "step": 231
+    },
+    {
+      "epoch": 0.8140350877192982,
+      "grad_norm": 0.8997906447720105,
+      "learning_rate": 1.565217391304348e-05,
+      "loss": 1.0236,
+      "step": 232
+    },
+    {
+      "epoch": 0.8175438596491228,
+      "grad_norm": 0.8927454629454755,
+      "learning_rate": 1.536231884057971e-05,
+      "loss": 0.96,
+      "step": 233
+    },
+    {
+      "epoch": 0.8210526315789474,
+      "grad_norm": 0.9150163082832309,
+      "learning_rate": 1.5072463768115944e-05,
+      "loss": 0.9843,
+      "step": 234
+    },
+    {
+      "epoch": 0.8245614035087719,
+      "grad_norm": 0.9497695522124332,
+      "learning_rate": 1.4782608695652174e-05,
+      "loss": 1.0254,
+      "step": 235
+    },
+    {
+      "epoch": 0.8280701754385965,
+      "grad_norm": 0.8134079237524361,
+      "learning_rate": 1.4492753623188407e-05,
+      "loss": 0.9874,
+      "step": 236
+    },
+    {
+      "epoch": 0.8315789473684211,
+      "grad_norm": 0.854532647109886,
+      "learning_rate": 1.420289855072464e-05,
+      "loss": 0.9697,
+      "step": 237
+    },
+    {
+      "epoch": 0.8350877192982457,
+      "grad_norm": 0.902401593136911,
+      "learning_rate": 1.391304347826087e-05,
+      "loss": 1.0083,
+      "step": 238
+    },
+    {
+      "epoch": 0.8385964912280702,
+      "grad_norm": 0.8365313967364223,
+      "learning_rate": 1.3623188405797103e-05,
+      "loss": 0.9479,
+      "step": 239
+    },
+    {
+      "epoch": 0.8421052631578947,
+      "grad_norm": 0.8385782446580461,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 0.998,
+      "step": 240
+    },
+    {
+      "epoch": 0.8456140350877193,
+      "grad_norm": 0.9130856203558422,
+      "learning_rate": 1.3043478260869566e-05,
+      "loss": 0.9934,
+      "step": 241
+    },
+    {
+      "epoch": 0.8491228070175438,
+      "grad_norm": 0.8261442785562587,
+      "learning_rate": 1.2753623188405797e-05,
+      "loss": 0.9274,
+      "step": 242
+    },
+    {
+      "epoch": 0.8526315789473684,
+      "grad_norm": 0.8560538840518993,
+      "learning_rate": 1.2463768115942029e-05,
+      "loss": 0.9511,
+      "step": 243
+    },
+    {
+      "epoch": 0.856140350877193,
+      "grad_norm": 0.8699104552871371,
+      "learning_rate": 1.2173913043478263e-05,
+      "loss": 0.958,
+      "step": 244
+    },
+    {
+      "epoch": 0.8596491228070176,
+      "grad_norm": 0.9107920631411749,
+      "learning_rate": 1.1884057971014494e-05,
+      "loss": 1.0177,
+      "step": 245
+    },
+    {
+      "epoch": 0.8631578947368421,
+      "grad_norm": 0.8776802277406592,
+      "learning_rate": 1.1594202898550726e-05,
+      "loss": 0.9261,
+      "step": 246
+    },
+    {
+      "epoch": 0.8666666666666667,
+      "grad_norm": 0.8999713914824392,
+      "learning_rate": 1.1304347826086957e-05,
+      "loss": 0.9973,
+      "step": 247
+    },
+    {
+      "epoch": 0.8701754385964913,
+      "grad_norm": 0.8835866950552989,
+      "learning_rate": 1.101449275362319e-05,
+      "loss": 1.0028,
+      "step": 248
+    },
+    {
+      "epoch": 0.8736842105263158,
+      "grad_norm": 0.9243563863707843,
+      "learning_rate": 1.0724637681159422e-05,
+      "loss": 0.9971,
+      "step": 249
+    },
+    {
+      "epoch": 0.8771929824561403,
+      "grad_norm": 0.9287603572835252,
+      "learning_rate": 1.0434782608695653e-05,
+      "loss": 0.9902,
+      "step": 250
+    },
+    {
+      "epoch": 0.8807017543859649,
+      "grad_norm": 0.8670790287211418,
+      "learning_rate": 1.0144927536231885e-05,
+      "loss": 1.0039,
+      "step": 251
+    },
+    {
+      "epoch": 0.8842105263157894,
+      "grad_norm": 0.8663791969051364,
+      "learning_rate": 9.855072463768118e-06,
+      "loss": 0.9817,
+      "step": 252
+    },
+    {
+      "epoch": 0.887719298245614,
+      "grad_norm": 0.8631015068740149,
+      "learning_rate": 9.565217391304349e-06,
+      "loss": 1.0173,
+      "step": 253
+    },
+    {
+      "epoch": 0.8912280701754386,
+      "grad_norm": 0.8829365467183184,
+      "learning_rate": 9.275362318840581e-06,
+      "loss": 0.9578,
+      "step": 254
+    },
+    {
+      "epoch": 0.8947368421052632,
+      "grad_norm": 0.9001038483145964,
+      "learning_rate": 8.985507246376812e-06,
+      "loss": 1.0258,
+      "step": 255
+    },
+    {
+      "epoch": 0.8982456140350877,
+      "grad_norm": 0.8834099073060356,
+      "learning_rate": 8.695652173913044e-06,
+      "loss": 0.9726,
+      "step": 256
+    },
+    {
+      "epoch": 0.9017543859649123,
+      "grad_norm": 0.8221574328830703,
+      "learning_rate": 8.405797101449275e-06,
+      "loss": 0.9318,
+      "step": 257
+    },
+    {
+      "epoch": 0.9052631578947369,
+      "grad_norm": 0.8603145991699224,
+      "learning_rate": 8.115942028985508e-06,
+      "loss": 0.9663,
+      "step": 258
+    },
+    {
+      "epoch": 0.9087719298245615,
+      "grad_norm": 0.8557514339905812,
+      "learning_rate": 7.82608695652174e-06,
+      "loss": 0.9968,
+      "step": 259
+    },
+    {
+      "epoch": 0.9122807017543859,
+      "grad_norm": 0.8263665060539463,
+      "learning_rate": 7.536231884057972e-06,
+      "loss": 1.0022,
+      "step": 260
+    },
+    {
+      "epoch": 0.9157894736842105,
+      "grad_norm": 0.9157072836324722,
+      "learning_rate": 7.246376811594203e-06,
+      "loss": 0.9387,
+      "step": 261
+    },
+    {
+      "epoch": 0.9192982456140351,
+      "grad_norm": 0.9233305451677062,
+      "learning_rate": 6.956521739130435e-06,
+      "loss": 0.9934,
+      "step": 262
+    },
+    {
+      "epoch": 0.9228070175438596,
+      "grad_norm": 0.903604456186341,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.9502,
+      "step": 263
+    },
+    {
+      "epoch": 0.9263157894736842,
+      "grad_norm": 0.9285076793745337,
+      "learning_rate": 6.376811594202898e-06,
+      "loss": 0.9645,
+      "step": 264
+    },
+    {
+      "epoch": 0.9298245614035088,
+      "grad_norm": 0.888909609341721,
+      "learning_rate": 6.086956521739132e-06,
+      "loss": 0.946,
+      "step": 265
+    },
+    {
+      "epoch": 0.9333333333333333,
+      "grad_norm": 0.9438455431857006,
+      "learning_rate": 5.797101449275363e-06,
+      "loss": 1.0153,
+      "step": 266
+    },
+    {
+      "epoch": 0.9368421052631579,
+      "grad_norm": 0.8551557910334344,
+      "learning_rate": 5.507246376811595e-06,
+      "loss": 0.958,
+      "step": 267
+    },
+    {
+      "epoch": 0.9403508771929825,
+      "grad_norm": 0.8234220234488231,
+      "learning_rate": 5.2173913043478265e-06,
+      "loss": 0.96,
+      "step": 268
+    },
+    {
+      "epoch": 0.9438596491228071,
+      "grad_norm": 0.9003874584417242,
+      "learning_rate": 4.927536231884059e-06,
+      "loss": 0.9757,
+      "step": 269
+    },
+    {
+      "epoch": 0.9473684210526315,
+      "grad_norm": 0.89021747241037,
+      "learning_rate": 4.637681159420291e-06,
+      "loss": 0.9511,
+      "step": 270
+    },
+    {
+      "epoch": 0.9508771929824561,
+      "grad_norm": 0.850549521859951,
+      "learning_rate": 4.347826086956522e-06,
+      "loss": 0.9619,
+      "step": 271
+    },
+    {
+      "epoch": 0.9543859649122807,
+      "grad_norm": 0.8477206031920781,
+      "learning_rate": 4.057971014492754e-06,
+      "loss": 0.9169,
+      "step": 272
+    },
+    {
+      "epoch": 0.9578947368421052,
+      "grad_norm": 0.9199733593357698,
+      "learning_rate": 3.768115942028986e-06,
+      "loss": 1.0299,
+      "step": 273
+    },
+    {
+      "epoch": 0.9614035087719298,
+      "grad_norm": 0.8038841148302012,
+      "learning_rate": 3.4782608695652175e-06,
+      "loss": 0.9584,
+      "step": 274
+    },
+    {
+      "epoch": 0.9649122807017544,
+      "grad_norm": 0.8488075839645941,
+      "learning_rate": 3.188405797101449e-06,
+      "loss": 0.9868,
+      "step": 275
+    },
+    {
+      "epoch": 0.968421052631579,
+      "grad_norm": 0.8355215697229909,
+      "learning_rate": 2.8985507246376816e-06,
+      "loss": 1.009,
+      "step": 276
+    },
+    {
+      "epoch": 0.9719298245614035,
+      "grad_norm": 0.9354156162463755,
+      "learning_rate": 2.6086956521739132e-06,
+      "loss": 0.9914,
+      "step": 277
+    },
+    {
+      "epoch": 0.9754385964912281,
+      "grad_norm": 0.8504047654484729,
+      "learning_rate": 2.3188405797101453e-06,
+      "loss": 1.0137,
+      "step": 278
+    },
+    {
+      "epoch": 0.9789473684210527,
+      "grad_norm": 0.9106574593705012,
+      "learning_rate": 2.028985507246377e-06,
+      "loss": 0.9963,
+      "step": 279
+    },
+    {
+      "epoch": 0.9824561403508771,
+      "grad_norm": 0.8356909679028329,
+      "learning_rate": 1.7391304347826088e-06,
+      "loss": 1.0176,
+      "step": 280
+    },
+    {
+      "epoch": 0.9859649122807017,
+      "grad_norm": 0.8904235043433295,
+      "learning_rate": 1.4492753623188408e-06,
+      "loss": 0.9684,
+      "step": 281
+    },
+    {
+      "epoch": 0.9894736842105263,
+      "grad_norm": 0.8663330662929399,
+      "learning_rate": 1.1594202898550726e-06,
+      "loss": 0.9852,
+      "step": 282
+    },
+    {
+      "epoch": 0.9929824561403509,
+      "grad_norm": 0.8453724734461978,
+      "learning_rate": 8.695652173913044e-07,
+      "loss": 0.9599,
+      "step": 283
+    },
+    {
+      "epoch": 0.9964912280701754,
+      "grad_norm": 0.8659094349837689,
+      "learning_rate": 5.797101449275363e-07,
+      "loss": 1.0253,
+      "step": 284
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.8132420261672163,
+      "learning_rate": 2.8985507246376816e-07,
+      "loss": 0.8957,
+      "step": 285
+    },
+    {
+      "epoch": 1.0,
+      "step": 285,
+      "total_flos": 224781354270720.0,
+      "train_loss": 1.0555618083267881,
+      "train_runtime": 2779.4288,
+      "train_samples_per_second": 52.454,
+      "train_steps_per_second": 0.103
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 285,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 700,
+  "total_flos": 224781354270720.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dca7fb36802253051575e9de9c44be43e98f54a18c8c686f622a8da95a7669a
+size 7096

zero_to_fp32.py

@@ -0,0 +1,604 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 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
+from dataclasses import dataclass
+
+# 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,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+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_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        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]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    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, exclude_frozen_parameters):
+    """
+    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_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # 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.")
+
+    # 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() if _has_callable(shape, 'numel') else math.prod(shape)
+            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")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    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 _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # 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
+
+    # 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
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # 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"Trainable params: {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 Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    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, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+    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("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)