initial commit: LexPodLM-13B

.gitattributes

@@ -32,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.psd filter=lfs diff=lfs merge=lfs -text

config.json

@@ -0,0 +1,24 @@
+{
+  "_name_or_path": "/workspace/datasets/llama-13b",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 2048,
+  "max_sequence_length": 2048,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "pad_token_id": 0,
+  "rms_norm_eps": 1e-06,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.28.1",
+  "use_cache": false,
+  "vocab_size": 32000
+}

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": "4.28.1"
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:316aa63bdb5b58c77dff5490f62fad85304252fa6c653cf389e13245ec4d7595
+size 26031865519

special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "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": "<unk>",
+  "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 sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,34 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "model_max_length": 2048,
+  "pad_token": null,
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

trainer_state.json

@@ -0,0 +1,1592 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.9941747572815534,
+  "global_step": 256,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 1.8181818181818183e-06,
+      "loss": 2.6279,
+      "step": 1
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.6363636363636366e-06,
+      "loss": 2.6631,
+      "step": 2
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.4545454545454545e-06,
+      "loss": 2.6104,
+      "step": 3
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 7.272727272727273e-06,
+      "loss": 2.6699,
+      "step": 4
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 2.5469,
+      "step": 5
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.0909090909090909e-05,
+      "loss": 2.499,
+      "step": 6
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.2727272727272728e-05,
+      "loss": 2.5264,
+      "step": 7
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.4545454545454546e-05,
+      "loss": 2.5,
+      "step": 8
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.6363636363636366e-05,
+      "loss": 2.5107,
+      "step": 9
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1.8181818181818182e-05,
+      "loss": 2.541,
+      "step": 10
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 2e-05,
+      "loss": 2.5381,
+      "step": 11
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1.9999177886783194e-05,
+      "loss": 2.5234,
+      "step": 12
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.99967116823068e-05,
+      "loss": 2.5303,
+      "step": 13
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1.999260179207068e-05,
+      "loss": 2.5117,
+      "step": 14
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9986848891833846e-05,
+      "loss": 2.4707,
+      "step": 15
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9979453927503366e-05,
+      "loss": 2.5107,
+      "step": 16
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.997041811497882e-05,
+      "loss": 2.5078,
+      "step": 17
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.9959742939952393e-05,
+      "loss": 2.4258,
+      "step": 18
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1.9947430157664575e-05,
+      "loss": 2.4717,
+      "step": 19
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.9933481792615583e-05,
+      "loss": 2.5234,
+      "step": 20
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.991790013823246e-05,
+      "loss": 2.4385,
+      "step": 21
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.9900687756492022e-05,
+      "loss": 2.4541,
+      "step": 22
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1.988184747749956e-05,
+      "loss": 2.4629,
+      "step": 23
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.986138239902355e-05,
+      "loss": 2.5381,
+      "step": 24
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.98392958859863e-05,
+      "loss": 2.4775,
+      "step": 25
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1.9815591569910654e-05,
+      "loss": 2.4717,
+      "step": 26
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.979027334832293e-05,
+      "loss": 2.4678,
+      "step": 27
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1.9763345384112044e-05,
+      "loss": 2.4863,
+      "step": 28
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.973481210484505e-05,
+      "loss": 2.5049,
+      "step": 29
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.9704678202039148e-05,
+      "loss": 2.5186,
+      "step": 30
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.9672948630390296e-05,
+      "loss": 2.3848,
+      "step": 31
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1.9639628606958535e-05,
+      "loss": 2.5127,
+      "step": 32
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.9604723610310195e-05,
+      "loss": 2.4727,
+      "step": 33
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.956823937961709e-05,
+      "loss": 2.4502,
+      "step": 34
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1.9530181913712875e-05,
+      "loss": 2.5029,
+      "step": 35
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.949055747010669e-05,
+      "loss": 2.5,
+      "step": 36
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.9449372563954293e-05,
+      "loss": 2.4551,
+      "step": 37
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.9406633966986828e-05,
+      "loss": 2.458,
+      "step": 38
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.9362348706397374e-05,
+      "loss": 2.4453,
+      "step": 39
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.9316524063685544e-05,
+      "loss": 2.3975,
+      "step": 40
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1.926916757346022e-05,
+      "loss": 2.4785,
+      "step": 41
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.9220287022200707e-05,
+      "loss": 2.416,
+      "step": 42
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.9169890446976454e-05,
+      "loss": 2.5166,
+      "step": 43
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.911798613412557e-05,
+      "loss": 2.4268,
+      "step": 44
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.9064582617892383e-05,
+      "loss": 2.4736,
+      "step": 45
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.900968867902419e-05,
+      "loss": 2.4531,
+      "step": 46
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.895331334332753e-05,
+      "loss": 2.4199,
+      "step": 47
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.889546588018412e-05,
+      "loss": 2.4893,
+      "step": 48
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.8836155801026754e-05,
+      "loss": 2.4785,
+      "step": 49
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1.877539285777543e-05,
+      "loss": 2.4346,
+      "step": 50
+    },
+    {
+      "epoch": 0.39,
+      "eval_loss": 2.4345414638519287,
+      "eval_runtime": 50.8104,
+      "eval_samples_per_second": 3.326,
+      "eval_steps_per_second": 0.039,
+      "step": 50
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.8713187041233896e-05,
+      "loss": 2.4541,
+      "step": 51
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.8649548579446938e-05,
+      "loss": 2.4746,
+      "step": 52
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.8584487936018663e-05,
+      "loss": 2.4355,
+      "step": 53
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.8518015808392045e-05,
+      "loss": 2.4268,
+      "step": 54
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.8450143126090015e-05,
+      "loss": 2.4092,
+      "step": 55
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.8380881048918406e-05,
+      "loss": 2.5039,
+      "step": 56
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.831024096513104e-05,
+      "loss": 2.4609,
+      "step": 57
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.8238234489557217e-05,
+      "loss": 2.5352,
+      "step": 58
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.8164873461691987e-05,
+      "loss": 2.4775,
+      "step": 59
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.8090169943749477e-05,
+      "loss": 2.4678,
+      "step": 60
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.8014136218679566e-05,
+      "loss": 2.4814,
+      "step": 61
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1.793678478814833e-05,
+      "loss": 2.4473,
+      "step": 62
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.7858128370482427e-05,
+      "loss": 2.4346,
+      "step": 63
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.7778179898577973e-05,
+      "loss": 2.4561,
+      "step": 64
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.769695251777406e-05,
+      "loss": 2.5137,
+      "step": 65
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.7614459583691346e-05,
+      "loss": 2.4717,
+      "step": 66
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.7530714660036112e-05,
+      "loss": 2.4561,
+      "step": 67
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.744573151637007e-05,
+      "loss": 2.5293,
+      "step": 68
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.7359524125846353e-05,
+      "loss": 2.4531,
+      "step": 69
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.7272106662911972e-05,
+      "loss": 2.4854,
+      "step": 70
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7183493500977277e-05,
+      "loss": 2.4414,
+      "step": 71
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.709369921005258e-05,
+      "loss": 2.4619,
+      "step": 72
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.700273855435255e-05,
+      "loss": 2.4424,
+      "step": 73
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.691062648986865e-05,
+      "loss": 2.4336,
+      "step": 74
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6817378161909995e-05,
+      "loss": 2.4795,
+      "step": 75
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.672300890261317e-05,
+      "loss": 2.4756,
+      "step": 76
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.662753422842123e-05,
+      "loss": 2.4492,
+      "step": 77
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.6530969837532487e-05,
+      "loss": 2.4141,
+      "step": 78
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.6433331607319342e-05,
+      "loss": 2.459,
+      "step": 79
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.63346355917177e-05,
+      "loss": 2.4434,
+      "step": 80
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.6234898018587336e-05,
+      "loss": 2.4854,
+      "step": 81
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.6134135287043668e-05,
+      "loss": 2.3984,
+      "step": 82
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.6032363964761363e-05,
+      "loss": 2.4277,
+      "step": 83
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.592960078525026e-05,
+      "loss": 2.415,
+      "step": 84
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.5825862645103962e-05,
+      "loss": 2.3994,
+      "step": 85
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.5721166601221697e-05,
+      "loss": 2.4014,
+      "step": 86
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.561552986800375e-05,
+      "loss": 2.4766,
+      "step": 87
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.5508969814521026e-05,
+      "loss": 2.4268,
+      "step": 88
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.5401503961659202e-05,
+      "loss": 2.4688,
+      "step": 89
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.5293149979237875e-05,
+      "loss": 2.4619,
+      "step": 90
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.5183925683105254e-05,
+      "loss": 2.4678,
+      "step": 91
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.5073849032208823e-05,
+      "loss": 2.4551,
+      "step": 92
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.4962938125642504e-05,
+      "loss": 2.4502,
+      "step": 93
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.485121119967072e-05,
+      "loss": 2.4707,
+      "step": 94
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.4738686624729987e-05,
+      "loss": 2.4453,
+      "step": 95
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.4625382902408356e-05,
+      "loss": 2.4443,
+      "step": 96
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.4511318662403347e-05,
+      "loss": 2.4424,
+      "step": 97
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1.4396512659458824e-05,
+      "loss": 2.4375,
+      "step": 98
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.4280983770281258e-05,
+      "loss": 2.4492,
+      "step": 99
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.4164750990435991e-05,
+      "loss": 2.3975,
+      "step": 100
+    },
+    {
+      "epoch": 0.78,
+      "eval_loss": 2.415125846862793,
+      "eval_runtime": 46.8913,
+      "eval_samples_per_second": 3.604,
+      "eval_steps_per_second": 0.043,
+      "step": 100
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.4047833431223938e-05,
+      "loss": 2.4238,
+      "step": 101
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.3930250316539237e-05,
+      "loss": 2.4658,
+      "step": 102
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.3812020979708418e-05,
+      "loss": 2.4521,
+      "step": 103
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.3693164860311565e-05,
+      "loss": 2.4619,
+      "step": 104
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.3573701500986012e-05,
+      "loss": 2.3975,
+      "step": 105
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.3453650544213078e-05,
+      "loss": 2.4551,
+      "step": 106
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1.333303172908842e-05,
+      "loss": 2.4492,
+      "step": 107
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.3211864888076458e-05,
+      "loss": 2.4092,
+      "step": 108
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.3090169943749475e-05,
+      "loss": 2.4395,
+      "step": 109
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.2967966905511906e-05,
+      "loss": 2.3906,
+      "step": 110
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.2845275866310325e-05,
+      "loss": 2.4619,
+      "step": 111
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1.2722116999329712e-05,
+      "loss": 2.4443,
+      "step": 112
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.259851055467653e-05,
+      "loss": 2.4629,
+      "step": 113
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.2474476856049145e-05,
+      "loss": 2.4502,
+      "step": 114
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.2350036297396153e-05,
+      "loss": 2.4736,
+      "step": 115
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.2225209339563144e-05,
+      "loss": 2.4521,
+      "step": 116
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.2100016506928494e-05,
+      "loss": 2.4775,
+      "step": 117
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1.1974478384028672e-05,
+      "loss": 2.3906,
+      "step": 118
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1.1848615612173689e-05,
+      "loss": 2.4521,
+      "step": 119
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.172244888605319e-05,
+      "loss": 2.4844,
+      "step": 120
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.1595998950333794e-05,
+      "loss": 2.4258,
+      "step": 121
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1.1469286596248181e-05,
+      "loss": 2.3906,
+      "step": 122
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.1342332658176556e-05,
+      "loss": 2.4238,
+      "step": 123
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.1215158010221005e-05,
+      "loss": 2.458,
+      "step": 124
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.108778356277331e-05,
+      "loss": 2.4736,
+      "step": 125
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.0960230259076819e-05,
+      "loss": 2.4062,
+      "step": 126
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.0832519071782895e-05,
+      "loss": 2.4004,
+      "step": 127
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.070467099950254e-05,
+      "loss": 2.4473,
+      "step": 128
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.0576707063353745e-05,
+      "loss": 2.3799,
+      "step": 129
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.044864830350515e-05,
+      "loss": 2.2168,
+      "step": 130
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.0320515775716556e-05,
+      "loss": 2.1523,
+      "step": 131
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 1.0192330547876871e-05,
+      "loss": 2.1826,
+      "step": 132
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 1.0064113696540112e-05,
+      "loss": 2.1543,
+      "step": 133
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 9.93588630345989e-06,
+      "loss": 2.1992,
+      "step": 134
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 9.80766945212313e-06,
+      "loss": 2.1069,
+      "step": 135
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 9.67948422428345e-06,
+      "loss": 2.1523,
+      "step": 136
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 9.551351696494854e-06,
+      "loss": 2.1797,
+      "step": 137
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 9.423292936646258e-06,
+      "loss": 2.1787,
+      "step": 138
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 9.29532900049746e-06,
+      "loss": 2.2017,
+      "step": 139
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 9.167480928217108e-06,
+      "loss": 2.1636,
+      "step": 140
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 9.039769740923183e-06,
+      "loss": 2.1738,
+      "step": 141
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 8.912216437226692e-06,
+      "loss": 2.1904,
+      "step": 142
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 8.784841989778997e-06,
+      "loss": 2.1855,
+      "step": 143
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 8.657667341823449e-06,
+      "loss": 2.1182,
+      "step": 144
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 8.530713403751822e-06,
+      "loss": 2.1924,
+      "step": 145
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 8.404001049666211e-06,
+      "loss": 2.1953,
+      "step": 146
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 8.277551113946812e-06,
+      "loss": 2.1802,
+      "step": 147
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 8.151384387826313e-06,
+      "loss": 2.1299,
+      "step": 148
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 8.02552161597133e-06,
+      "loss": 2.166,
+      "step": 149
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 7.899983493071506e-06,
+      "loss": 2.1167,
+      "step": 150
+    },
+    {
+      "epoch": 1.17,
+      "eval_loss": 2.4375,
+      "eval_runtime": 44.0835,
+      "eval_samples_per_second": 3.834,
+      "eval_steps_per_second": 0.045,
+      "step": 150
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 7.774790660436857e-06,
+      "loss": 2.1729,
+      "step": 151
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 7.649963702603848e-06,
+      "loss": 2.1719,
+      "step": 152
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 7.525523143950859e-06,
+      "loss": 2.2061,
+      "step": 153
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 7.401489445323473e-06,
+      "loss": 2.1426,
+      "step": 154
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 7.27788300067029e-06,
+      "loss": 2.1987,
+      "step": 155
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 7.154724133689677e-06,
+      "loss": 2.1572,
+      "step": 156
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 7.032033094488094e-06,
+      "loss": 2.1465,
+      "step": 157
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 6.909830056250527e-06,
+      "loss": 2.1411,
+      "step": 158
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 6.788135111923545e-06,
+      "loss": 2.1548,
+      "step": 159
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 6.666968270911585e-06,
+      "loss": 2.1138,
+      "step": 160
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 6.546349455786926e-06,
+      "loss": 2.1548,
+      "step": 161
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 6.426298499013994e-06,
+      "loss": 2.1782,
+      "step": 162
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 6.306835139688439e-06,
+      "loss": 2.1621,
+      "step": 163
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 6.187979020291584e-06,
+      "loss": 2.1738,
+      "step": 164
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 6.069749683460765e-06,
+      "loss": 2.1221,
+      "step": 165
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 5.952166568776062e-06,
+      "loss": 2.1304,
+      "step": 166
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 5.835249009564013e-06,
+      "loss": 2.1885,
+      "step": 167
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 5.719016229718748e-06,
+      "loss": 2.1777,
+      "step": 168
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 5.60348734054118e-06,
+      "loss": 2.1304,
+      "step": 169
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 5.488681337596653e-06,
+      "loss": 2.1294,
+      "step": 170
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 5.37461709759165e-06,
+      "loss": 2.1592,
+      "step": 171
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 5.2613133752700145e-06,
+      "loss": 2.1367,
+      "step": 172
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 5.148788800329279e-06,
+      "loss": 2.124,
+      "step": 173
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 5.037061874357503e-06,
+      "loss": 2.1538,
+      "step": 174
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 4.92615096779118e-06,
+      "loss": 2.1382,
+      "step": 175
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 4.81607431689475e-06,
+      "loss": 2.1826,
+      "step": 176
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 4.706850020762126e-06,
+      "loss": 2.167,
+      "step": 177
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 4.598496038340801e-06,
+      "loss": 2.1699,
+      "step": 178
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 4.491030185478976e-06,
+      "loss": 2.1533,
+      "step": 179
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 4.3844701319962525e-06,
+      "loss": 2.1255,
+      "step": 180
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 4.278833398778306e-06,
+      "loss": 2.1577,
+      "step": 181
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 4.17413735489604e-06,
+      "loss": 2.1519,
+      "step": 182
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 4.070399214749743e-06,
+      "loss": 2.1636,
+      "step": 183
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 3.967636035238636e-06,
+      "loss": 2.1699,
+      "step": 184
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 3.865864712956336e-06,
+      "loss": 2.1719,
+      "step": 185
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 3.7651019814126656e-06,
+      "loss": 2.1538,
+      "step": 186
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 3.665364408282305e-06,
+      "loss": 2.2119,
+      "step": 187
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 3.5666683926806623e-06,
+      "loss": 2.1455,
+      "step": 188
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 3.4690301624675127e-06,
+      "loss": 2.1162,
+      "step": 189
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 3.372465771578771e-06,
+      "loss": 2.0981,
+      "step": 190
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 3.2769910973868314e-06,
+      "loss": 2.1162,
+      "step": 191
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 3.1826218380900066e-06,
+      "loss": 2.1655,
+      "step": 192
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 3.089373510131354e-06,
+      "loss": 2.1724,
+      "step": 193
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 2.9972614456474537e-06,
+      "loss": 2.0962,
+      "step": 194
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 2.9063007899474214e-06,
+      "loss": 2.1387,
+      "step": 195
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 2.8165064990227255e-06,
+      "loss": 2.1211,
+      "step": 196
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 2.7278933370880267e-06,
+      "loss": 2.1543,
+      "step": 197
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 2.640475874153651e-06,
+      "loss": 2.1494,
+      "step": 198
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 2.5542684836299316e-06,
+      "loss": 2.208,
+      "step": 199
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 2.469285339963892e-06,
+      "loss": 2.1606,
+      "step": 200
+    },
+    {
+      "epoch": 1.55,
+      "eval_loss": 2.4366679191589355,
+      "eval_runtime": 44.9926,
+      "eval_samples_per_second": 3.756,
+      "eval_steps_per_second": 0.044,
+      "step": 200
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 2.3855404163086558e-06,
+      "loss": 2.1528,
+      "step": 201
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 2.3030474822259396e-06,
+      "loss": 2.1606,
+      "step": 202
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 2.2218201014220266e-06,
+      "loss": 2.1572,
+      "step": 203
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 2.1418716295175766e-06,
+      "loss": 2.1582,
+      "step": 204
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 2.063215211851678e-06,
+      "loss": 2.1406,
+      "step": 205
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 1.9858637813204352e-06,
+      "loss": 2.1709,
+      "step": 206
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 1.9098300562505266e-06,
+      "loss": 2.1826,
+      "step": 207
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 1.835126538308013e-06,
+      "loss": 2.1416,
+      "step": 208
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 1.7617655104427833e-06,
+      "loss": 2.1665,
+      "step": 209
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 1.6897590348689607e-06,
+      "loss": 2.1582,
+      "step": 210
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 1.6191189510815942e-06,
+      "loss": 2.144,
+      "step": 211
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.5498568739099907e-06,
+      "loss": 2.1685,
+      "step": 212
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.481984191607959e-06,
+      "loss": 2.1343,
+      "step": 213
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 1.4155120639813392e-06,
+      "loss": 2.1528,
+      "step": 214
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 1.350451420553065e-06,
+      "loss": 2.1411,
+      "step": 215
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 1.286812958766106e-06,
+      "loss": 2.1289,
+      "step": 216
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 1.224607142224572e-06,
+      "loss": 2.127,
+      "step": 217
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 1.1638441989732474e-06,
+      "loss": 2.1367,
+      "step": 218
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 1.1045341198158833e-06,
+      "loss": 2.1021,
+      "step": 219
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 1.0466866566724698e-06,
+      "loss": 2.1436,
+      "step": 220
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 9.903113209758098e-07,
+      "loss": 2.1509,
+      "step": 221
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 9.354173821076184e-07,
+      "loss": 2.1665,
+      "step": 222
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 8.820138658744304e-07,
+      "loss": 2.1162,
+      "step": 223
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 8.301095530235492e-07,
+      "loss": 2.1104,
+      "step": 224
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 7.797129777992951e-07,
+      "loss": 2.1348,
+      "step": 225
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 7.308324265397837e-07,
+      "loss": 2.1562,
+      "step": 226
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 6.834759363144595e-07,
+      "loss": 2.1543,
+      "step": 227
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 6.37651293602628e-07,
+      "loss": 2.1357,
+      "step": 228
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 5.933660330131752e-07,
+      "loss": 2.1392,
+      "step": 229
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 5.506274360457087e-07,
+      "loss": 2.123,
+      "step": 230
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 5.094425298933136e-07,
+      "loss": 2.1606,
+      "step": 231
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 4.6981808628712823e-07,
+      "loss": 2.0991,
+      "step": 232
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 4.3176062038291275e-07,
+      "loss": 2.1582,
+      "step": 233
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 3.9527638968980707e-07,
+      "loss": 2.1929,
+      "step": 234
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 3.603713930414676e-07,
+      "loss": 2.0874,
+      "step": 235
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 3.2705136960970554e-07,
+      "loss": 2.1299,
+      "step": 236
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 2.9532179796085356e-07,
+      "loss": 2.1704,
+      "step": 237
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 2.6518789515495356e-07,
+      "loss": 2.1382,
+      "step": 238
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 2.3665461588795902e-07,
+      "loss": 2.1689,
+      "step": 239
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 2.097266516770713e-07,
+      "loss": 2.1436,
+      "step": 240
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 1.844084300893456e-07,
+      "loss": 2.1108,
+      "step": 241
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 1.6070411401370335e-07,
+      "loss": 2.1206,
+      "step": 242
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 1.3861760097645062e-07,
+      "loss": 2.1118,
+      "step": 243
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 1.1815252250044318e-07,
+      "loss": 2.1738,
+      "step": 244
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 9.931224350798185e-08,
+      "loss": 2.1328,
+      "step": 245
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 8.209986176753947e-08,
+      "loss": 2.1538,
+      "step": 246
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 6.65182073844195e-08,
+      "loss": 2.1338,
+      "step": 247
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 5.256984233542595e-08,
+      "loss": 2.1455,
+      "step": 248
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 4.025706004760932e-08,
+      "loss": 2.1831,
+      "step": 249
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 2.9581885021181534e-08,
+      "loss": 2.1304,
+      "step": 250
+    },
+    {
+      "epoch": 1.94,
+      "eval_loss": 2.4366679191589355,
+      "eval_runtime": 40.7598,
+      "eval_samples_per_second": 4.146,
+      "eval_steps_per_second": 0.049,
+      "step": 250
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 2.054607249663665e-08,
+      "loss": 2.1387,
+      "step": 251
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 1.3151108166156168e-08,
+      "loss": 2.1406,
+      "step": 252
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 7.3982079293233314e-09,
+      "loss": 2.1309,
+      "step": 253
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 3.2883176932019255e-09,
+      "loss": 2.1133,
+      "step": 254
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 8.221132168073631e-10,
+      "loss": 2.1543,
+      "step": 255
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.0,
+      "loss": 2.1475,
+      "step": 256
+    }
+  ],
+  "max_steps": 256,
+  "num_train_epochs": 2,
+  "total_flos": 83249552424960.0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a95bb91ac445ca680f147f9c52cedc230507eaf729b4551416a08d2430966a5
+size 4591

zero_to_fp32.py

@@ -0,0 +1,578 @@
+#!/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 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_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_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)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+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 _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()
+            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):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _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):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _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):
+    """
+    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)