Upload folder using huggingface_hub

138_128_e3_3e-5/adapter_config.json

@@ -0,0 +1,39 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "ibm-granite/granite-3.3-8b-base",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "v_proj",
+    "gate_proj",
+    "up_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_rslora": false
+}

138_128_e3_3e-5/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3b40562a396a106d5ace370eeb5fc676f001331053fa2963c050a0473025f1c
+size 791751704

138_128_e3_3e-5/latest

@@ -0,0 +1 @@
+global_step1206

138_128_e3_3e-5/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e088c209b71004e477e8f2add844f86a36baa26e6e2559a2e2235e341e4e82bd
+size 16389

138_128_e3_3e-5/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69a4669c7440b0fc4af94a6032c09ab670348a8d5297e9fd12d87a550ae2547f
+size 16389

138_128_e3_3e-5/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ac3ccc7fe50337b39fbd42e0fb2243a8677bd59c5881912b8bc85a992a742f2
+size 16389

138_128_e3_3e-5/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba30b23484981c7a139c4aa29ca9290b00999e16427d52276edd3e4b0296e5af
+size 16389

138_128_e3_3e-5/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5467239d13215dc27600c254399639615000aa026115d03b760ba8ae28bc7d55
+size 16389

138_128_e3_3e-5/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc3af686e290920a438825f38130a0145a52d5de87552cccd4ffb2b74d33e01d
+size 16389

138_128_e3_3e-5/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:723ce1f386908a67bc367f294f663560c66c7ccf9bf8fae3acf42dda4f61d255
+size 16389

138_128_e3_3e-5/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:036dff1e5d0f36f8ffc0085eaf65a01e5391fbdc0479049440899b787ee7a5d2
+size 16389

138_128_e3_3e-5/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c15cbd142e69709606504fa4915f2c35e10aac34ede6b3eea570a384d000e192
+size 1401

138_128_e3_3e-5/special_tokens_map.json

@@ -0,0 +1,45 @@
+{
+  "additional_special_tokens": [
+    "<|endoftext|>",
+    "<fim_prefix>",
+    "<fim_middle>",
+    "<fim_suffix>",
+    "<fim_pad>",
+    "<filename>",
+    "<gh_stars>",
+    "<issue_start>",
+    "<issue_comment>",
+    "<issue_closed>",
+    "<jupyter_start>",
+    "<jupyter_text>",
+    "<jupyter_code>",
+    "<jupyter_output>",
+    "<empty_output>",
+    "<commit_before>",
+    "<commit_msg>",
+    "<commit_after>",
+    "<reponame>"
+  ],
+  "bos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<reponame>",
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

138_128_e3_3e-5/tokenizer_config.json

@@ -0,0 +1,188 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<fim_prefix>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<fim_middle>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<fim_suffix>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "<fim_pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "5": {
+      "content": "<filename>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "6": {
+      "content": "<gh_stars>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "7": {
+      "content": "<issue_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "8": {
+      "content": "<issue_comment>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "9": {
+      "content": "<issue_closed>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "10": {
+      "content": "<jupyter_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "11": {
+      "content": "<jupyter_text>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "12": {
+      "content": "<jupyter_code>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "13": {
+      "content": "<jupyter_output>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "14": {
+      "content": "<empty_output>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "15": {
+      "content": "<commit_before>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "16": {
+      "content": "<commit_msg>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "17": {
+      "content": "<commit_after>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "18": {
+      "content": "<reponame>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|endoftext|>",
+    "<fim_prefix>",
+    "<fim_middle>",
+    "<fim_suffix>",
+    "<fim_pad>",
+    "<filename>",
+    "<gh_stars>",
+    "<issue_start>",
+    "<issue_comment>",
+    "<issue_closed>",
+    "<jupyter_start>",
+    "<jupyter_text>",
+    "<jupyter_code>",
+    "<jupyter_output>",
+    "<empty_output>",
+    "<commit_before>",
+    "<commit_msg>",
+    "<commit_after>",
+    "<reponame>"
+  ],
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "extra_special_tokens": {},
+  "model_max_length": 8192,
+  "pad_token": "<reponame>",
+  "padding_side": "left",
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>",
+  "vocab_size": 49152
+}

138_128_e3_3e-5/trainer_state.json

@@ -0,0 +1,1721 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 1206,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.012453300124533,
+      "grad_norm": 0.8080592751502991,
+      "learning_rate": 1.9672131147540985e-06,
+      "loss": 1.3588,
+      "step": 5
+    },
+    {
+      "epoch": 0.024906600249066,
+      "grad_norm": 0.8234268426895142,
+      "learning_rate": 4.426229508196722e-06,
+      "loss": 1.3824,
+      "step": 10
+    },
+    {
+      "epoch": 0.037359900373599,
+      "grad_norm": 0.6236813068389893,
+      "learning_rate": 6.885245901639345e-06,
+      "loss": 1.3449,
+      "step": 15
+    },
+    {
+      "epoch": 0.049813200498132,
+      "grad_norm": 0.7997186779975891,
+      "learning_rate": 9.344262295081968e-06,
+      "loss": 1.3622,
+      "step": 20
+    },
+    {
+      "epoch": 0.062266500622665005,
+      "grad_norm": 0.5782414674758911,
+      "learning_rate": 1.180327868852459e-05,
+      "loss": 1.3563,
+      "step": 25
+    },
+    {
+      "epoch": 0.074719800747198,
+      "grad_norm": 0.5041614770889282,
+      "learning_rate": 1.4262295081967213e-05,
+      "loss": 1.275,
+      "step": 30
+    },
+    {
+      "epoch": 0.08717310087173101,
+      "grad_norm": 0.5814838409423828,
+      "learning_rate": 1.6721311475409834e-05,
+      "loss": 1.2895,
+      "step": 35
+    },
+    {
+      "epoch": 0.099626400996264,
+      "grad_norm": 0.445229172706604,
+      "learning_rate": 1.9180327868852462e-05,
+      "loss": 1.257,
+      "step": 40
+    },
+    {
+      "epoch": 0.11207970112079702,
+      "grad_norm": 0.5867952108383179,
+      "learning_rate": 2.1639344262295084e-05,
+      "loss": 1.2939,
+      "step": 45
+    },
+    {
+      "epoch": 0.12453300124533001,
+      "grad_norm": 0.5468295216560364,
+      "learning_rate": 2.4098360655737705e-05,
+      "loss": 1.3136,
+      "step": 50
+    },
+    {
+      "epoch": 0.136986301369863,
+      "grad_norm": 0.5516449809074402,
+      "learning_rate": 2.6557377049180327e-05,
+      "loss": 1.167,
+      "step": 55
+    },
+    {
+      "epoch": 0.149439601494396,
+      "grad_norm": 0.6059405207633972,
+      "learning_rate": 2.901639344262295e-05,
+      "loss": 1.2424,
+      "step": 60
+    },
+    {
+      "epoch": 0.16189290161892902,
+      "grad_norm": 0.5352672934532166,
+      "learning_rate": 2.9999491852149543e-05,
+      "loss": 1.1928,
+      "step": 65
+    },
+    {
+      "epoch": 0.17434620174346202,
+      "grad_norm": 0.5617313981056213,
+      "learning_rate": 2.999638662885322e-05,
+      "loss": 1.1933,
+      "step": 70
+    },
+    {
+      "epoch": 0.18679950186799502,
+      "grad_norm": 0.7198558449745178,
+      "learning_rate": 2.9990459070319718e-05,
+      "loss": 1.1228,
+      "step": 75
+    },
+    {
+      "epoch": 0.199252801992528,
+      "grad_norm": 0.6769135594367981,
+      "learning_rate": 2.9981710292121587e-05,
+      "loss": 1.1695,
+      "step": 80
+    },
+    {
+      "epoch": 0.21170610211706103,
+      "grad_norm": 0.6060968637466431,
+      "learning_rate": 2.9970141940787794e-05,
+      "loss": 1.0684,
+      "step": 85
+    },
+    {
+      "epoch": 0.22415940224159403,
+      "grad_norm": 0.6032590270042419,
+      "learning_rate": 2.9955756193493843e-05,
+      "loss": 1.129,
+      "step": 90
+    },
+    {
+      "epoch": 0.23661270236612703,
+      "grad_norm": 0.7026798725128174,
+      "learning_rate": 2.9938555757652027e-05,
+      "loss": 1.1037,
+      "step": 95
+    },
+    {
+      "epoch": 0.24906600249066002,
+      "grad_norm": 0.6857986450195312,
+      "learning_rate": 2.991854387040189e-05,
+      "loss": 1.0722,
+      "step": 100
+    },
+    {
+      "epoch": 0.261519302615193,
+      "grad_norm": 0.798175036907196,
+      "learning_rate": 2.9895724298000995e-05,
+      "loss": 0.9799,
+      "step": 105
+    },
+    {
+      "epoch": 0.273972602739726,
+      "grad_norm": 0.7469162940979004,
+      "learning_rate": 2.9870101335116107e-05,
+      "loss": 1.0434,
+      "step": 110
+    },
+    {
+      "epoch": 0.286425902864259,
+      "grad_norm": 0.8346182107925415,
+      "learning_rate": 2.9841679804014938e-05,
+      "loss": 1.0192,
+      "step": 115
+    },
+    {
+      "epoch": 0.298879202988792,
+      "grad_norm": 0.8008555173873901,
+      "learning_rate": 2.981046505365859e-05,
+      "loss": 1.0091,
+      "step": 120
+    },
+    {
+      "epoch": 0.31133250311332505,
+      "grad_norm": 0.737636923789978,
+      "learning_rate": 2.9776462958694873e-05,
+      "loss": 0.9642,
+      "step": 125
+    },
+    {
+      "epoch": 0.32378580323785805,
+      "grad_norm": 0.793342113494873,
+      "learning_rate": 2.9739679918352686e-05,
+      "loss": 0.9396,
+      "step": 130
+    },
+    {
+      "epoch": 0.33623910336239105,
+      "grad_norm": 0.9377651810646057,
+      "learning_rate": 2.9700122855237685e-05,
+      "loss": 0.9496,
+      "step": 135
+    },
+    {
+      "epoch": 0.34869240348692404,
+      "grad_norm": 0.8817457556724548,
+      "learning_rate": 2.965779921402944e-05,
+      "loss": 0.9266,
+      "step": 140
+    },
+    {
+      "epoch": 0.36114570361145704,
+      "grad_norm": 1.0338335037231445,
+      "learning_rate": 2.961271696008033e-05,
+      "loss": 0.8602,
+      "step": 145
+    },
+    {
+      "epoch": 0.37359900373599003,
+      "grad_norm": 1.0290571451187134,
+      "learning_rate": 2.9564884577916463e-05,
+      "loss": 0.9652,
+      "step": 150
+    },
+    {
+      "epoch": 0.386052303860523,
+      "grad_norm": 1.050189733505249,
+      "learning_rate": 2.951431106964088e-05,
+      "loss": 0.8365,
+      "step": 155
+    },
+    {
+      "epoch": 0.398505603985056,
+      "grad_norm": 0.946395218372345,
+      "learning_rate": 2.9461005953239347e-05,
+      "loss": 0.878,
+      "step": 160
+    },
+    {
+      "epoch": 0.410958904109589,
+      "grad_norm": 0.9408192038536072,
+      "learning_rate": 2.9404979260789064e-05,
+      "loss": 0.8259,
+      "step": 165
+    },
+    {
+      "epoch": 0.42341220423412207,
+      "grad_norm": 1.0667990446090698,
+      "learning_rate": 2.934624153657061e-05,
+      "loss": 0.8414,
+      "step": 170
+    },
+    {
+      "epoch": 0.43586550435865506,
+      "grad_norm": 1.0747874975204468,
+      "learning_rate": 2.9284803835083507e-05,
+      "loss": 0.8448,
+      "step": 175
+    },
+    {
+      "epoch": 0.44831880448318806,
+      "grad_norm": 1.1501070261001587,
+      "learning_rate": 2.9220677718965747e-05,
+      "loss": 0.7738,
+      "step": 180
+    },
+    {
+      "epoch": 0.46077210460772106,
+      "grad_norm": 1.1795527935028076,
+      "learning_rate": 2.9153875256817696e-05,
+      "loss": 0.793,
+      "step": 185
+    },
+    {
+      "epoch": 0.47322540473225405,
+      "grad_norm": 1.2069406509399414,
+      "learning_rate": 2.9084409020930767e-05,
+      "loss": 0.8194,
+      "step": 190
+    },
+    {
+      "epoch": 0.48567870485678705,
+      "grad_norm": 1.017179250717163,
+      "learning_rate": 2.9012292084921306e-05,
+      "loss": 0.7806,
+      "step": 195
+    },
+    {
+      "epoch": 0.49813200498132004,
+      "grad_norm": 1.210204005241394,
+      "learning_rate": 2.893753802127012e-05,
+      "loss": 0.717,
+      "step": 200
+    },
+    {
+      "epoch": 0.5105853051058531,
+      "grad_norm": 1.050508737564087,
+      "learning_rate": 2.8860160898768123e-05,
+      "loss": 0.7412,
+      "step": 205
+    },
+    {
+      "epoch": 0.523038605230386,
+      "grad_norm": 1.113065481185913,
+      "learning_rate": 2.8780175279868577e-05,
+      "loss": 0.7257,
+      "step": 210
+    },
+    {
+      "epoch": 0.5354919053549191,
+      "grad_norm": 1.277635931968689,
+      "learning_rate": 2.8697596217946426e-05,
+      "loss": 0.6946,
+      "step": 215
+    },
+    {
+      "epoch": 0.547945205479452,
+      "grad_norm": 1.13856840133667,
+      "learning_rate": 2.861243925446523e-05,
+      "loss": 0.6996,
+      "step": 220
+    },
+    {
+      "epoch": 0.5603985056039851,
+      "grad_norm": 1.1662951707839966,
+      "learning_rate": 2.8524720416052243e-05,
+      "loss": 0.7071,
+      "step": 225
+    },
+    {
+      "epoch": 0.572851805728518,
+      "grad_norm": 1.255119800567627,
+      "learning_rate": 2.84344562114822e-05,
+      "loss": 0.6925,
+      "step": 230
+    },
+    {
+      "epoch": 0.5853051058530511,
+      "grad_norm": 1.0429599285125732,
+      "learning_rate": 2.8341663628570328e-05,
+      "loss": 0.6548,
+      "step": 235
+    },
+    {
+      "epoch": 0.597758405977584,
+      "grad_norm": 1.1847997903823853,
+      "learning_rate": 2.824636013097524e-05,
+      "loss": 0.6159,
+      "step": 240
+    },
+    {
+      "epoch": 0.6102117061021171,
+      "grad_norm": 1.3273192644119263,
+      "learning_rate": 2.8148563654912257e-05,
+      "loss": 0.6803,
+      "step": 245
+    },
+    {
+      "epoch": 0.6226650062266501,
+      "grad_norm": 1.2407212257385254,
+      "learning_rate": 2.8048292605777766e-05,
+      "loss": 0.6159,
+      "step": 250
+    },
+    {
+      "epoch": 0.635118306351183,
+      "grad_norm": 1.393981695175171,
+      "learning_rate": 2.7945565854685348e-05,
+      "loss": 0.6562,
+      "step": 255
+    },
+    {
+      "epoch": 0.6475716064757161,
+      "grad_norm": 1.354434847831726,
+      "learning_rate": 2.7840402734914182e-05,
+      "loss": 0.5831,
+      "step": 260
+    },
+    {
+      "epoch": 0.660024906600249,
+      "grad_norm": 1.4470314979553223,
+      "learning_rate": 2.773282303827052e-05,
+      "loss": 0.6376,
+      "step": 265
+    },
+    {
+      "epoch": 0.6724782067247821,
+      "grad_norm": 1.4742989540100098,
+      "learning_rate": 2.762284701136283e-05,
+      "loss": 0.6223,
+      "step": 270
+    },
+    {
+      "epoch": 0.684931506849315,
+      "grad_norm": 1.4081107378005981,
+      "learning_rate": 2.7510495351791397e-05,
+      "loss": 0.6028,
+      "step": 275
+    },
+    {
+      "epoch": 0.6973848069738481,
+      "grad_norm": 1.5004246234893799,
+      "learning_rate": 2.739578920425297e-05,
+      "loss": 0.623,
+      "step": 280
+    },
+    {
+      "epoch": 0.709838107098381,
+      "grad_norm": 1.3481271266937256,
+      "learning_rate": 2.727875015656135e-05,
+      "loss": 0.5735,
+      "step": 285
+    },
+    {
+      "epoch": 0.7222914072229141,
+      "grad_norm": 1.2337456941604614,
+      "learning_rate": 2.7159400235584507e-05,
+      "loss": 0.563,
+      "step": 290
+    },
+    {
+      "epoch": 0.7347447073474471,
+      "grad_norm": 1.4883575439453125,
+      "learning_rate": 2.703776190309914e-05,
+      "loss": 0.5608,
+      "step": 295
+    },
+    {
+      "epoch": 0.7471980074719801,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 2.691385805156329e-05,
+      "loss": 0.6026,
+      "step": 300
+    },
+    {
+      "epoch": 0.7596513075965131,
+      "grad_norm": 1.2968716621398926,
+      "learning_rate": 2.6787711999808026e-05,
+      "loss": 0.5371,
+      "step": 305
+    },
+    {
+      "epoch": 0.772104607721046,
+      "grad_norm": 1.4366308450698853,
+      "learning_rate": 2.6659347488648763e-05,
+      "loss": 0.538,
+      "step": 310
+    },
+    {
+      "epoch": 0.7845579078455791,
+      "grad_norm": 1.4500861167907715,
+      "learning_rate": 2.6528788676417238e-05,
+      "loss": 0.4939,
+      "step": 315
+    },
+    {
+      "epoch": 0.797011207970112,
+      "grad_norm": 1.4863909482955933,
+      "learning_rate": 2.6396060134414883e-05,
+      "loss": 0.5191,
+      "step": 320
+    },
+    {
+      "epoch": 0.8094645080946451,
+      "grad_norm": 1.4849021434783936,
+      "learning_rate": 2.6261186842288482e-05,
+      "loss": 0.4927,
+      "step": 325
+    },
+    {
+      "epoch": 0.821917808219178,
+      "grad_norm": 1.4660779237747192,
+      "learning_rate": 2.6124194183328992e-05,
+      "loss": 0.5037,
+      "step": 330
+    },
+    {
+      "epoch": 0.8343711083437111,
+      "grad_norm": 2.036705493927002,
+      "learning_rate": 2.5985107939694346e-05,
+      "loss": 0.5341,
+      "step": 335
+    },
+    {
+      "epoch": 0.8468244084682441,
+      "grad_norm": 1.3688981533050537,
+      "learning_rate": 2.5843954287557253e-05,
+      "loss": 0.4643,
+      "step": 340
+    },
+    {
+      "epoch": 0.8592777085927771,
+      "grad_norm": 1.3126728534698486,
+      "learning_rate": 2.5700759792178813e-05,
+      "loss": 0.5322,
+      "step": 345
+    },
+    {
+      "epoch": 0.8717310087173101,
+      "grad_norm": 1.3128783702850342,
+      "learning_rate": 2.5555551402908896e-05,
+      "loss": 0.4773,
+      "step": 350
+    },
+    {
+      "epoch": 0.8841843088418431,
+      "grad_norm": 1.2541402578353882,
+      "learning_rate": 2.5408356448114255e-05,
+      "loss": 0.4621,
+      "step": 355
+    },
+    {
+      "epoch": 0.8966376089663761,
+      "grad_norm": 1.583799123764038,
+      "learning_rate": 2.5259202630035296e-05,
+      "loss": 0.4391,
+      "step": 360
+    },
+    {
+      "epoch": 0.9090909090909091,
+      "grad_norm": 1.4803394079208374,
+      "learning_rate": 2.51081180195725e-05,
+      "loss": 0.4523,
+      "step": 365
+    },
+    {
+      "epoch": 0.9215442092154421,
+      "grad_norm": 1.2685974836349487,
+      "learning_rate": 2.4955131051003427e-05,
+      "loss": 0.4585,
+      "step": 370
+    },
+    {
+      "epoch": 0.933997509339975,
+      "grad_norm": 1.3847997188568115,
+      "learning_rate": 2.4800270516631376e-05,
+      "loss": 0.4191,
+      "step": 375
+    },
+    {
+      "epoch": 0.9464508094645081,
+      "grad_norm": 1.3490678071975708,
+      "learning_rate": 2.4643565561366644e-05,
+      "loss": 0.4432,
+      "step": 380
+    },
+    {
+      "epoch": 0.958904109589041,
+      "grad_norm": 1.434613585472107,
+      "learning_rate": 2.4485045677241415e-05,
+      "loss": 0.4085,
+      "step": 385
+    },
+    {
+      "epoch": 0.9713574097135741,
+      "grad_norm": 1.4781972169876099,
+      "learning_rate": 2.4324740697859326e-05,
+      "loss": 0.4609,
+      "step": 390
+    },
+    {
+      "epoch": 0.9838107098381071,
+      "grad_norm": 1.4535627365112305,
+      "learning_rate": 2.4162680792780775e-05,
+      "loss": 0.4156,
+      "step": 395
+    },
+    {
+      "epoch": 0.9962640099626401,
+      "grad_norm": 1.2948943376541138,
+      "learning_rate": 2.399889646184494e-05,
+      "loss": 0.4055,
+      "step": 400
+    },
+    {
+      "epoch": 1.0074719800747198,
+      "grad_norm": 1.420023798942566,
+      "learning_rate": 2.3833418529429728e-05,
+      "loss": 0.351,
+      "step": 405
+    },
+    {
+      "epoch": 1.0199252801992529,
+      "grad_norm": 1.250096321105957,
+      "learning_rate": 2.366627813865055e-05,
+      "loss": 0.2938,
+      "step": 410
+    },
+    {
+      "epoch": 1.0323785803237857,
+      "grad_norm": 1.5623716115951538,
+      "learning_rate": 2.349750674549918e-05,
+      "loss": 0.3404,
+      "step": 415
+    },
+    {
+      "epoch": 1.0448318804483188,
+      "grad_norm": 1.1456372737884521,
+      "learning_rate": 2.332713611292371e-05,
+      "loss": 0.3281,
+      "step": 420
+    },
+    {
+      "epoch": 1.0572851805728518,
+      "grad_norm": 1.3642221689224243,
+      "learning_rate": 2.3155198304850694e-05,
+      "loss": 0.3128,
+      "step": 425
+    },
+    {
+      "epoch": 1.0697384806973849,
+      "grad_norm": 1.3247578144073486,
+      "learning_rate": 2.2981725680150745e-05,
+      "loss": 0.3057,
+      "step": 430
+    },
+    {
+      "epoch": 1.0821917808219177,
+      "grad_norm": 1.447908878326416,
+      "learning_rate": 2.2806750886548508e-05,
+      "loss": 0.3395,
+      "step": 435
+    },
+    {
+      "epoch": 1.0946450809464507,
+      "grad_norm": 1.375496745109558,
+      "learning_rate": 2.2630306854478335e-05,
+      "loss": 0.3347,
+      "step": 440
+    },
+    {
+      "epoch": 1.1070983810709838,
+      "grad_norm": 1.3178277015686035,
+      "learning_rate": 2.245242679088679e-05,
+      "loss": 0.3132,
+      "step": 445
+    },
+    {
+      "epoch": 1.1195516811955168,
+      "grad_norm": 1.8294039964675903,
+      "learning_rate": 2.2273144172982985e-05,
+      "loss": 0.3204,
+      "step": 450
+    },
+    {
+      "epoch": 1.13200498132005,
+      "grad_norm": 1.3418508768081665,
+      "learning_rate": 2.2092492741938222e-05,
+      "loss": 0.3207,
+      "step": 455
+    },
+    {
+      "epoch": 1.1444582814445827,
+      "grad_norm": 1.8415858745574951,
+      "learning_rate": 2.1910506496535816e-05,
+      "loss": 0.278,
+      "step": 460
+    },
+    {
+      "epoch": 1.1569115815691158,
+      "grad_norm": 1.214003562927246,
+      "learning_rate": 2.1727219686772494e-05,
+      "loss": 0.294,
+      "step": 465
+    },
+    {
+      "epoch": 1.1693648816936488,
+      "grad_norm": 1.2681173086166382,
+      "learning_rate": 2.154266680741253e-05,
+      "loss": 0.3033,
+      "step": 470
+    },
+    {
+      "epoch": 1.1818181818181819,
+      "grad_norm": 1.4074482917785645,
+      "learning_rate": 2.1356882591495795e-05,
+      "loss": 0.2956,
+      "step": 475
+    },
+    {
+      "epoch": 1.1942714819427147,
+      "grad_norm": 1.342892050743103,
+      "learning_rate": 2.116990200380093e-05,
+      "loss": 0.3226,
+      "step": 480
+    },
+    {
+      "epoch": 1.2067247820672478,
+      "grad_norm": 1.498038649559021,
+      "learning_rate": 2.0981760234264983e-05,
+      "loss": 0.2925,
+      "step": 485
+    },
+    {
+      "epoch": 1.2191780821917808,
+      "grad_norm": 1.4801772832870483,
+      "learning_rate": 2.07924926913606e-05,
+      "loss": 0.2973,
+      "step": 490
+    },
+    {
+      "epoch": 1.2316313823163139,
+      "grad_norm": 1.6219295263290405,
+      "learning_rate": 2.0602134995432124e-05,
+      "loss": 0.236,
+      "step": 495
+    },
+    {
+      "epoch": 1.244084682440847,
+      "grad_norm": 1.4914093017578125,
+      "learning_rate": 2.0410722971991802e-05,
+      "loss": 0.2865,
+      "step": 500
+    },
+    {
+      "epoch": 1.25653798256538,
+      "grad_norm": 1.211521029472351,
+      "learning_rate": 2.0218292644977396e-05,
+      "loss": 0.2717,
+      "step": 505
+    },
+    {
+      "epoch": 1.2689912826899128,
+      "grad_norm": 1.4810200929641724,
+      "learning_rate": 2.002488022997244e-05,
+      "loss": 0.2754,
+      "step": 510
+    },
+    {
+      "epoch": 1.2814445828144458,
+      "grad_norm": 1.3870760202407837,
+      "learning_rate": 1.9830522127390428e-05,
+      "loss": 0.2579,
+      "step": 515
+    },
+    {
+      "epoch": 1.293897882938979,
+      "grad_norm": 1.4146753549575806,
+      "learning_rate": 1.963525491562421e-05,
+      "loss": 0.2647,
+      "step": 520
+    },
+    {
+      "epoch": 1.3063511830635117,
+      "grad_norm": 1.2141958475112915,
+      "learning_rate": 1.943911534416193e-05,
+      "loss": 0.2617,
+      "step": 525
+    },
+    {
+      "epoch": 1.3188044831880448,
+      "grad_norm": 1.6414520740509033,
+      "learning_rate": 1.924214032667069e-05,
+      "loss": 0.246,
+      "step": 530
+    },
+    {
+      "epoch": 1.3312577833125778,
+      "grad_norm": 1.3524872064590454,
+      "learning_rate": 1.9044366934049408e-05,
+      "loss": 0.2937,
+      "step": 535
+    },
+    {
+      "epoch": 1.3437110834371109,
+      "grad_norm": 1.349511981010437,
+      "learning_rate": 1.8845832387451995e-05,
+      "loss": 0.2314,
+      "step": 540
+    },
+    {
+      "epoch": 1.356164383561644,
+      "grad_norm": 1.498072624206543,
+      "learning_rate": 1.8646574051282337e-05,
+      "loss": 0.2485,
+      "step": 545
+    },
+    {
+      "epoch": 1.3686176836861768,
+      "grad_norm": 1.256630539894104,
+      "learning_rate": 1.844662942616224e-05,
+      "loss": 0.2388,
+      "step": 550
+    },
+    {
+      "epoch": 1.3810709838107098,
+      "grad_norm": 1.382404088973999,
+      "learning_rate": 1.8246036141873786e-05,
+      "loss": 0.2434,
+      "step": 555
+    },
+    {
+      "epoch": 1.3935242839352429,
+      "grad_norm": 1.4739421606063843,
+      "learning_rate": 1.804483195027739e-05,
+      "loss": 0.2391,
+      "step": 560
+    },
+    {
+      "epoch": 1.405977584059776,
+      "grad_norm": 1.340551733970642,
+      "learning_rate": 1.7843054718206818e-05,
+      "loss": 0.2056,
+      "step": 565
+    },
+    {
+      "epoch": 1.4184308841843087,
+      "grad_norm": 1.3163232803344727,
+      "learning_rate": 1.7640742420342672e-05,
+      "loss": 0.2503,
+      "step": 570
+    },
+    {
+      "epoch": 1.4308841843088418,
+      "grad_norm": 1.4177124500274658,
+      "learning_rate": 1.7437933132065452e-05,
+      "loss": 0.2534,
+      "step": 575
+    },
+    {
+      "epoch": 1.4433374844333748,
+      "grad_norm": 1.3499191999435425,
+      "learning_rate": 1.7234665022289777e-05,
+      "loss": 0.2403,
+      "step": 580
+    },
+    {
+      "epoch": 1.455790784557908,
+      "grad_norm": 1.5210360288619995,
+      "learning_rate": 1.7030976346280924e-05,
+      "loss": 0.2166,
+      "step": 585
+    },
+    {
+      "epoch": 1.468244084682441,
+      "grad_norm": 1.3385121822357178,
+      "learning_rate": 1.6826905438455174e-05,
+      "loss": 0.2148,
+      "step": 590
+    },
+    {
+      "epoch": 1.4806973848069738,
+      "grad_norm": 1.3507510423660278,
+      "learning_rate": 1.662249070516523e-05,
+      "loss": 0.2206,
+      "step": 595
+    },
+    {
+      "epoch": 1.4931506849315068,
+      "grad_norm": 1.3014428615570068,
+      "learning_rate": 1.641777061747209e-05,
+      "loss": 0.2041,
+      "step": 600
+    },
+    {
+      "epoch": 1.5056039850560399,
+      "grad_norm": 1.2394065856933594,
+      "learning_rate": 1.621278370390476e-05,
+      "loss": 0.2163,
+      "step": 605
+    },
+    {
+      "epoch": 1.5180572851805727,
+      "grad_norm": 1.4289287328720093,
+      "learning_rate": 1.6007568543209153e-05,
+      "loss": 0.207,
+      "step": 610
+    },
+    {
+      "epoch": 1.5305105853051058,
+      "grad_norm": 1.3962420225143433,
+      "learning_rate": 1.5802163757087513e-05,
+      "loss": 0.199,
+      "step": 615
+    },
+    {
+      "epoch": 1.5429638854296388,
+      "grad_norm": 1.3510708808898926,
+      "learning_rate": 1.5596608002929793e-05,
+      "loss": 0.1848,
+      "step": 620
+    },
+    {
+      "epoch": 1.5554171855541719,
+      "grad_norm": 1.3102413415908813,
+      "learning_rate": 1.539093996653829e-05,
+      "loss": 0.2047,
+      "step": 625
+    },
+    {
+      "epoch": 1.567870485678705,
+      "grad_norm": 1.247463583946228,
+      "learning_rate": 1.518519835484691e-05,
+      "loss": 0.1933,
+      "step": 630
+    },
+    {
+      "epoch": 1.580323785803238,
+      "grad_norm": 1.442697286605835,
+      "learning_rate": 1.4979421888636532e-05,
+      "loss": 0.1965,
+      "step": 635
+    },
+    {
+      "epoch": 1.592777085927771,
+      "grad_norm": 1.2049447298049927,
+      "learning_rate": 1.4773649295247668e-05,
+      "loss": 0.2078,
+      "step": 640
+    },
+    {
+      "epoch": 1.6052303860523038,
+      "grad_norm": 1.3384487628936768,
+      "learning_rate": 1.4567919301291976e-05,
+      "loss": 0.208,
+      "step": 645
+    },
+    {
+      "epoch": 1.6176836861768369,
+      "grad_norm": 1.2717690467834473,
+      "learning_rate": 1.4362270625363852e-05,
+      "loss": 0.1924,
+      "step": 650
+    },
+    {
+      "epoch": 1.6301369863013697,
+      "grad_norm": 1.292299747467041,
+      "learning_rate": 1.415674197075355e-05,
+      "loss": 0.1971,
+      "step": 655
+    },
+    {
+      "epoch": 1.6425902864259028,
+      "grad_norm": 1.2976208925247192,
+      "learning_rate": 1.3951372018163197e-05,
+      "loss": 0.1701,
+      "step": 660
+    },
+    {
+      "epoch": 1.6550435865504358,
+      "grad_norm": 1.4109532833099365,
+      "learning_rate": 1.3746199418427044e-05,
+      "loss": 0.1876,
+      "step": 665
+    },
+    {
+      "epoch": 1.6674968866749689,
+      "grad_norm": 1.1893404722213745,
+      "learning_rate": 1.3541262785237321e-05,
+      "loss": 0.1964,
+      "step": 670
+    },
+    {
+      "epoch": 1.679950186799502,
+      "grad_norm": 1.3668186664581299,
+      "learning_rate": 1.3336600687877124e-05,
+      "loss": 0.176,
+      "step": 675
+    },
+    {
+      "epoch": 1.692403486924035,
+      "grad_norm": 1.1720973253250122,
+      "learning_rate": 1.313225164396162e-05,
+      "loss": 0.1792,
+      "step": 680
+    },
+    {
+      "epoch": 1.704856787048568,
+      "grad_norm": 1.2465754747390747,
+      "learning_rate": 1.2928254112189e-05,
+      "loss": 0.1794,
+      "step": 685
+    },
+    {
+      "epoch": 1.7173100871731009,
+      "grad_norm": 1.4014456272125244,
+      "learning_rate": 1.272464648510251e-05,
+      "loss": 0.1553,
+      "step": 690
+    },
+    {
+      "epoch": 1.729763387297634,
+      "grad_norm": 1.3747260570526123,
+      "learning_rate": 1.2521467081864945e-05,
+      "loss": 0.1661,
+      "step": 695
+    },
+    {
+      "epoch": 1.7422166874221667,
+      "grad_norm": 1.320011854171753,
+      "learning_rate": 1.2318754141046936e-05,
+      "loss": 0.1616,
+      "step": 700
+    },
+    {
+      "epoch": 1.7546699875466998,
+      "grad_norm": 1.2350207567214966,
+      "learning_rate": 1.211654581343039e-05,
+      "loss": 0.1632,
+      "step": 705
+    },
+    {
+      "epoch": 1.7671232876712328,
+      "grad_norm": 1.2444283962249756,
+      "learning_rate": 1.1914880154828514e-05,
+      "loss": 0.1487,
+      "step": 710
+    },
+    {
+      "epoch": 1.7795765877957659,
+      "grad_norm": 1.5010842084884644,
+      "learning_rate": 1.1713795118923659e-05,
+      "loss": 0.18,
+      "step": 715
+    },
+    {
+      "epoch": 1.792029887920299,
+      "grad_norm": 1.1842529773712158,
+      "learning_rate": 1.1513328550124379e-05,
+      "loss": 0.1512,
+      "step": 720
+    },
+    {
+      "epoch": 1.804483188044832,
+      "grad_norm": 1.143868088722229,
+      "learning_rate": 1.1313518176443099e-05,
+      "loss": 0.1506,
+      "step": 725
+    },
+    {
+      "epoch": 1.816936488169365,
+      "grad_norm": 1.1031876802444458,
+      "learning_rate": 1.1114401602395647e-05,
+      "loss": 0.1642,
+      "step": 730
+    },
+    {
+      "epoch": 1.8293897882938979,
+      "grad_norm": 1.1954771280288696,
+      "learning_rate": 1.0916016301924056e-05,
+      "loss": 0.1727,
+      "step": 735
+    },
+    {
+      "epoch": 1.841843088418431,
+      "grad_norm": 1.185741662979126,
+      "learning_rate": 1.071839961134393e-05,
+      "loss": 0.1435,
+      "step": 740
+    },
+    {
+      "epoch": 1.8542963885429637,
+      "grad_norm": 1.1315739154815674,
+      "learning_rate": 1.0521588722317707e-05,
+      "loss": 0.1617,
+      "step": 745
+    },
+    {
+      "epoch": 1.8667496886674968,
+      "grad_norm": 1.292269229888916,
+      "learning_rate": 1.0325620674855147e-05,
+      "loss": 0.1535,
+      "step": 750
+    },
+    {
+      "epoch": 1.8792029887920298,
+      "grad_norm": 1.203553557395935,
+      "learning_rate": 1.0130532350342381e-05,
+      "loss": 0.1304,
+      "step": 755
+    },
+    {
+      "epoch": 1.891656288916563,
+      "grad_norm": 1.0999270677566528,
+      "learning_rate": 9.936360464600769e-06,
+      "loss": 0.1494,
+      "step": 760
+    },
+    {
+      "epoch": 1.904109589041096,
+      "grad_norm": 1.1319379806518555,
+      "learning_rate": 9.74314156097697e-06,
+      "loss": 0.1312,
+      "step": 765
+    },
+    {
+      "epoch": 1.916562889165629,
+      "grad_norm": 1.2186167240142822,
+      "learning_rate": 9.550912003465442e-06,
+      "loss": 0.1397,
+      "step": 770
+    },
+    {
+      "epoch": 1.929016189290162,
+      "grad_norm": 1.2208855152130127,
+      "learning_rate": 9.359707969864688e-06,
+      "loss": 0.1309,
+      "step": 775
+    },
+    {
+      "epoch": 1.9414694894146949,
+      "grad_norm": 1.074632167816162,
+      "learning_rate": 9.16956544496857e-06,
+      "loss": 0.138,
+      "step": 780
+    },
+    {
+      "epoch": 1.953922789539228,
+      "grad_norm": 1.2034265995025635,
+      "learning_rate": 8.980520213793934e-06,
+      "loss": 0.1505,
+      "step": 785
+    },
+    {
+      "epoch": 1.9663760896637608,
+      "grad_norm": 1.1010161638259888,
+      "learning_rate": 8.792607854845829e-06,
+      "loss": 0.1301,
+      "step": 790
+    },
+    {
+      "epoch": 1.9788293897882938,
+      "grad_norm": 1.102501392364502,
+      "learning_rate": 8.605863733421594e-06,
+      "loss": 0.1299,
+      "step": 795
+    },
+    {
+      "epoch": 1.9912826899128269,
+      "grad_norm": 1.2707922458648682,
+      "learning_rate": 8.420322994955074e-06,
+      "loss": 0.1257,
+      "step": 800
+    },
+    {
+      "epoch": 2.0024906600249066,
+      "grad_norm": 1.0714095830917358,
+      "learning_rate": 8.236020558402222e-06,
+      "loss": 0.12,
+      "step": 805
+    },
+    {
+      "epoch": 2.0149439601494397,
+      "grad_norm": 0.9427981376647949,
+      "learning_rate": 8.052991109669306e-06,
+      "loss": 0.0977,
+      "step": 810
+    },
+    {
+      "epoch": 2.0273972602739727,
+      "grad_norm": 0.9927377700805664,
+      "learning_rate": 7.87126909508499e-06,
+      "loss": 0.1054,
+      "step": 815
+    },
+    {
+      "epoch": 2.0398505603985058,
+      "grad_norm": 1.0380443334579468,
+      "learning_rate": 7.690888714917507e-06,
+      "loss": 0.1114,
+      "step": 820
+    },
+    {
+      "epoch": 2.052303860523039,
+      "grad_norm": 1.1310601234436035,
+      "learning_rate": 7.511883916938109e-06,
+      "loss": 0.1012,
+      "step": 825
+    },
+    {
+      "epoch": 2.0647571606475714,
+      "grad_norm": 0.9412599802017212,
+      "learning_rate": 7.334288390032098e-06,
+      "loss": 0.1026,
+      "step": 830
+    },
+    {
+      "epoch": 2.0772104607721045,
+      "grad_norm": 1.0357398986816406,
+      "learning_rate": 7.158135557858515e-06,
+      "loss": 0.0989,
+      "step": 835
+    },
+    {
+      "epoch": 2.0896637608966375,
+      "grad_norm": 1.0631839036941528,
+      "learning_rate": 6.983458572559782e-06,
+      "loss": 0.102,
+      "step": 840
+    },
+    {
+      "epoch": 2.1021170610211706,
+      "grad_norm": 0.8926840424537659,
+      "learning_rate": 6.81029030852244e-06,
+      "loss": 0.09,
+      "step": 845
+    },
+    {
+      "epoch": 2.1145703611457036,
+      "grad_norm": 0.8600242137908936,
+      "learning_rate": 6.63866335619015e-06,
+      "loss": 0.0933,
+      "step": 850
+    },
+    {
+      "epoch": 2.1270236612702367,
+      "grad_norm": 0.9629788398742676,
+      "learning_rate": 6.468610015930143e-06,
+      "loss": 0.0933,
+      "step": 855
+    },
+    {
+      "epoch": 2.1394769613947697,
+      "grad_norm": 0.9406184554100037,
+      "learning_rate": 6.3001622919542495e-06,
+      "loss": 0.1018,
+      "step": 860
+    },
+    {
+      "epoch": 2.151930261519303,
+      "grad_norm": 0.9864522814750671,
+      "learning_rate": 6.133351886295691e-06,
+      "loss": 0.0997,
+      "step": 865
+    },
+    {
+      "epoch": 2.1643835616438354,
+      "grad_norm": 0.8337077498435974,
+      "learning_rate": 5.9682101928426966e-06,
+      "loss": 0.0905,
+      "step": 870
+    },
+    {
+      "epoch": 2.1768368617683684,
+      "grad_norm": 1.0092023611068726,
+      "learning_rate": 5.804768291430174e-06,
+      "loss": 0.1024,
+      "step": 875
+    },
+    {
+      "epoch": 2.1892901618929015,
+      "grad_norm": 0.8762721419334412,
+      "learning_rate": 5.643056941990433e-06,
+      "loss": 0.0878,
+      "step": 880
+    },
+    {
+      "epoch": 2.2017434620174345,
+      "grad_norm": 1.088156819343567,
+      "learning_rate": 5.483106578764136e-06,
+      "loss": 0.0916,
+      "step": 885
+    },
+    {
+      "epoch": 2.2141967621419676,
+      "grad_norm": 1.1185767650604248,
+      "learning_rate": 5.324947304572553e-06,
+      "loss": 0.0976,
+      "step": 890
+    },
+    {
+      "epoch": 2.2266500622665006,
+      "grad_norm": 1.003543734550476,
+      "learning_rate": 5.1686088851521685e-06,
+      "loss": 0.0928,
+      "step": 895
+    },
+    {
+      "epoch": 2.2391033623910337,
+      "grad_norm": 0.7943710088729858,
+      "learning_rate": 5.014120743552749e-06,
+      "loss": 0.093,
+      "step": 900
+    },
+    {
+      "epoch": 2.2515566625155667,
+      "grad_norm": 1.0742517709732056,
+      "learning_rate": 4.861511954599883e-06,
+      "loss": 0.0902,
+      "step": 905
+    },
+    {
+      "epoch": 2.2640099626401,
+      "grad_norm": 1.3119049072265625,
+      "learning_rate": 4.710811239423083e-06,
+      "loss": 0.0867,
+      "step": 910
+    },
+    {
+      "epoch": 2.276463262764633,
+      "grad_norm": 0.9398558735847473,
+      "learning_rate": 4.5620469600504355e-06,
+      "loss": 0.0869,
+      "step": 915
+    },
+    {
+      "epoch": 2.2889165628891655,
+      "grad_norm": 0.8181595802307129,
+      "learning_rate": 4.415247114070834e-06,
+      "loss": 0.082,
+      "step": 920
+    },
+    {
+      "epoch": 2.3013698630136985,
+      "grad_norm": 0.7236878871917725,
+      "learning_rate": 4.270439329364799e-06,
+      "loss": 0.0797,
+      "step": 925
+    },
+    {
+      "epoch": 2.3138231631382316,
+      "grad_norm": 1.060689091682434,
+      "learning_rate": 4.1276508589048986e-06,
+      "loss": 0.0831,
+      "step": 930
+    },
+    {
+      "epoch": 2.3262764632627646,
+      "grad_norm": 0.8227666616439819,
+      "learning_rate": 3.986908575626699e-06,
+      "loss": 0.0923,
+      "step": 935
+    },
+    {
+      "epoch": 2.3387297633872977,
+      "grad_norm": 1.6744259595870972,
+      "learning_rate": 3.848238967371265e-06,
+      "loss": 0.0797,
+      "step": 940
+    },
+    {
+      "epoch": 2.3511830635118307,
+      "grad_norm": 0.7763772010803223,
+      "learning_rate": 3.7116681319001018e-06,
+      "loss": 0.0814,
+      "step": 945
+    },
+    {
+      "epoch": 2.3636363636363638,
+      "grad_norm": 0.9393787384033203,
+      "learning_rate": 3.5772217719835384e-06,
+      "loss": 0.0924,
+      "step": 950
+    },
+    {
+      "epoch": 2.376089663760897,
+      "grad_norm": 0.7159603834152222,
+      "learning_rate": 3.444925190563445e-06,
+      "loss": 0.0738,
+      "step": 955
+    },
+    {
+      "epoch": 2.3885429638854294,
+      "grad_norm": 0.7752528190612793,
+      "learning_rate": 3.3148032859911844e-06,
+      "loss": 0.0848,
+      "step": 960
+    },
+    {
+      "epoch": 2.4009962640099625,
+      "grad_norm": 0.8927635550498962,
+      "learning_rate": 3.186880547341727e-06,
+      "loss": 0.0837,
+      "step": 965
+    },
+    {
+      "epoch": 2.4134495641344955,
+      "grad_norm": 0.909500241279602,
+      "learning_rate": 3.0611810498047742e-06,
+      "loss": 0.0909,
+      "step": 970
+    },
+    {
+      "epoch": 2.4259028642590286,
+      "grad_norm": 0.7461646199226379,
+      "learning_rate": 2.937728450153789e-06,
+      "loss": 0.0833,
+      "step": 975
+    },
+    {
+      "epoch": 2.4383561643835616,
+      "grad_norm": 0.9491526484489441,
+      "learning_rate": 2.816545982293752e-06,
+      "loss": 0.0783,
+      "step": 980
+    },
+    {
+      "epoch": 2.4508094645080947,
+      "grad_norm": 0.8411048054695129,
+      "learning_rate": 2.6976564528885422e-06,
+      "loss": 0.0778,
+      "step": 985
+    },
+    {
+      "epoch": 2.4632627646326277,
+      "grad_norm": 0.8202574253082275,
+      "learning_rate": 2.5810822370686804e-06,
+      "loss": 0.0794,
+      "step": 990
+    },
+    {
+      "epoch": 2.4757160647571608,
+      "grad_norm": 0.7827599048614502,
+      "learning_rate": 2.466845274220316e-06,
+      "loss": 0.0828,
+      "step": 995
+    },
+    {
+      "epoch": 2.488169364881694,
+      "grad_norm": 0.8397166132926941,
+      "learning_rate": 2.3549670638562016e-06,
+      "loss": 0.0717,
+      "step": 1000
+    },
+    {
+      "epoch": 2.500622665006227,
+      "grad_norm": 0.7251439094543457,
+      "learning_rate": 2.2454686615694785e-06,
+      "loss": 0.0771,
+      "step": 1005
+    },
+    {
+      "epoch": 2.51307596513076,
+      "grad_norm": 0.7172680497169495,
+      "learning_rate": 2.138370675070977e-06,
+      "loss": 0.0844,
+      "step": 1010
+    },
+    {
+      "epoch": 2.5255292652552925,
+      "grad_norm": 0.8408495783805847,
+      "learning_rate": 2.0336932603108355e-06,
+      "loss": 0.0756,
+      "step": 1015
+    },
+    {
+      "epoch": 2.5379825653798256,
+      "grad_norm": 0.7621785402297974,
+      "learning_rate": 1.9314561176851235e-06,
+      "loss": 0.0817,
+      "step": 1020
+    },
+    {
+      "epoch": 2.5504358655043586,
+      "grad_norm": 0.7755963206291199,
+      "learning_rate": 1.8316784883282105e-06,
+      "loss": 0.0803,
+      "step": 1025
+    },
+    {
+      "epoch": 2.5628891656288917,
+      "grad_norm": 0.7943703532218933,
+      "learning_rate": 1.7343791504915684e-06,
+      "loss": 0.0821,
+      "step": 1030
+    },
+    {
+      "epoch": 2.5753424657534247,
+      "grad_norm": 0.8613129258155823,
+      "learning_rate": 1.6395764160096678e-06,
+      "loss": 0.0794,
+      "step": 1035
+    },
+    {
+      "epoch": 2.587795765877958,
+      "grad_norm": 0.875903844833374,
+      "learning_rate": 1.547288126853697e-06,
+      "loss": 0.0794,
+      "step": 1040
+    },
+    {
+      "epoch": 2.6002490660024904,
+      "grad_norm": 0.7708663940429688,
+      "learning_rate": 1.4575316517736714e-06,
+      "loss": 0.0811,
+      "step": 1045
+    },
+    {
+      "epoch": 2.6127023661270234,
+      "grad_norm": 0.9445556998252869,
+      "learning_rate": 1.370323883029615e-06,
+      "loss": 0.0855,
+      "step": 1050
+    },
+    {
+      "epoch": 2.6251556662515565,
+      "grad_norm": 0.8990893959999084,
+      "learning_rate": 1.2856812332124274e-06,
+      "loss": 0.0743,
+      "step": 1055
+    },
+    {
+      "epoch": 2.6376089663760895,
+      "grad_norm": 0.6993798017501831,
+      "learning_rate": 1.2036196321550096e-06,
+      "loss": 0.0675,
+      "step": 1060
+    },
+    {
+      "epoch": 2.6500622665006226,
+      "grad_norm": 0.7997357249259949,
+      "learning_rate": 1.1241545239342609e-06,
+      "loss": 0.0777,
+      "step": 1065
+    },
+    {
+      "epoch": 2.6625155666251556,
+      "grad_norm": 0.743100643157959,
+      "learning_rate": 1.0473008639644814e-06,
+      "loss": 0.0681,
+      "step": 1070
+    },
+    {
+      "epoch": 2.6749688667496887,
+      "grad_norm": 0.650947630405426,
+      "learning_rate": 9.730731161827528e-07,
+      "loss": 0.0701,
+      "step": 1075
+    },
+    {
+      "epoch": 2.6874221668742218,
+      "grad_norm": 0.9549784064292908,
+      "learning_rate": 9.014852503268045e-07,
+      "loss": 0.0715,
+      "step": 1080
+    },
+    {
+      "epoch": 2.699875466998755,
+      "grad_norm": 0.7748307585716248,
+      "learning_rate": 8.325507393059101e-07,
+      "loss": 0.0645,
+      "step": 1085
+    },
+    {
+      "epoch": 2.712328767123288,
+      "grad_norm": 0.7659729719161987,
+      "learning_rate": 7.662825566652442e-07,
+      "loss": 0.0731,
+      "step": 1090
+    },
+    {
+      "epoch": 2.724782067247821,
+      "grad_norm": 0.8021079897880554,
+      "learning_rate": 7.026931741442783e-07,
+      "loss": 0.0798,
+      "step": 1095
+    },
+    {
+      "epoch": 2.7372353673723535,
+      "grad_norm": 0.7025024890899658,
+      "learning_rate": 6.417945593295638e-07,
+      "loss": 0.0797,
+      "step": 1100
+    },
+    {
+      "epoch": 2.7496886674968866,
+      "grad_norm": 0.9227949976921082,
+      "learning_rate": 5.835981734024348e-07,
+      "loss": 0.083,
+      "step": 1105
+    },
+    {
+      "epoch": 2.7621419676214196,
+      "grad_norm": 0.7839409708976746,
+      "learning_rate": 5.281149689819981e-07,
+      "loss": 0.0753,
+      "step": 1110
+    },
+    {
+      "epoch": 2.7745952677459527,
+      "grad_norm": 0.8874260783195496,
+      "learning_rate": 4.7535538806383006e-07,
+      "loss": 0.0748,
+      "step": 1115
+    },
+    {
+      "epoch": 2.7870485678704857,
+      "grad_norm": 0.6409869194030762,
+      "learning_rate": 4.2532936005479585e-07,
+      "loss": 0.0727,
+      "step": 1120
+    },
+    {
+      "epoch": 2.7995018679950188,
+      "grad_norm": 0.727016806602478,
+      "learning_rate": 3.7804629990431884e-07,
+      "loss": 0.0686,
+      "step": 1125
+    },
+    {
+      "epoch": 2.811955168119552,
+      "grad_norm": 0.8032270669937134,
+      "learning_rate": 3.335151063324765e-07,
+      "loss": 0.0739,
+      "step": 1130
+    },
+    {
+      "epoch": 2.8244084682440844,
+      "grad_norm": 0.614982008934021,
+      "learning_rate": 2.917441601552534e-07,
+      "loss": 0.068,
+      "step": 1135
+    },
+    {
+      "epoch": 2.8368617683686175,
+      "grad_norm": 0.8577175736427307,
+      "learning_rate": 2.527413227072628e-07,
+      "loss": 0.0843,
+      "step": 1140
+    },
+    {
+      "epoch": 2.8493150684931505,
+      "grad_norm": 0.7702206373214722,
+      "learning_rate": 2.165139343622352e-07,
+      "loss": 0.0866,
+      "step": 1145
+    },
+    {
+      "epoch": 2.8617683686176836,
+      "grad_norm": 0.7843850255012512,
+      "learning_rate": 1.830688131515551e-07,
+      "loss": 0.0792,
+      "step": 1150
+    },
+    {
+      "epoch": 2.8742216687422166,
+      "grad_norm": 0.7210635542869568,
+      "learning_rate": 1.5241225348109898e-07,
+      "loss": 0.0753,
+      "step": 1155
+    },
+    {
+      "epoch": 2.8866749688667497,
+      "grad_norm": 1.0072154998779297,
+      "learning_rate": 1.2455002494661972e-07,
+      "loss": 0.0819,
+      "step": 1160
+    },
+    {
+      "epoch": 2.8991282689912827,
+      "grad_norm": 0.7139269113540649,
+      "learning_rate": 9.948737124790331e-08,
+      "loss": 0.0712,
+      "step": 1165
+    },
+    {
+      "epoch": 2.911581569115816,
+      "grad_norm": 0.807988166809082,
+      "learning_rate": 7.722900920190179e-08,
+      "loss": 0.0712,
+      "step": 1170
+    },
+    {
+      "epoch": 2.924034869240349,
+      "grad_norm": 0.6790947318077087,
+      "learning_rate": 5.777912785502493e-08,
+      "loss": 0.0733,
+      "step": 1175
+    },
+    {
+      "epoch": 2.936488169364882,
+      "grad_norm": 0.7935905456542969,
+      "learning_rate": 4.114138769474918e-08,
+      "loss": 0.0788,
+      "step": 1180
+    },
+    {
+      "epoch": 2.948941469489415,
+      "grad_norm": 0.7174673676490784,
+      "learning_rate": 2.731891996071878e-08,
+      "loss": 0.0729,
+      "step": 1185
+    },
+    {
+      "epoch": 2.9613947696139475,
+      "grad_norm": 0.728678822517395,
+      "learning_rate": 1.6314326055440475e-08,
+      "loss": 0.073,
+      "step": 1190
+    },
+    {
+      "epoch": 2.9738480697384806,
+      "grad_norm": 0.786389946937561,
+      "learning_rate": 8.129677054693474e-09,
+      "loss": 0.0726,
+      "step": 1195
+    },
+    {
+      "epoch": 2.9863013698630136,
+      "grad_norm": 0.7849173545837402,
+      "learning_rate": 2.7665133177545708e-09,
+      "loss": 0.0696,
+      "step": 1200
+    },
+    {
+      "epoch": 2.9987546699875467,
+      "grad_norm": 0.7936521172523499,
+      "learning_rate": 2.2584419750504293e-10,
+      "loss": 0.0836,
+      "step": 1205
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 1206,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 2000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.541926523517272e+18,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

138_128_e3_3e-5/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4045c297feb94fb5a48f65ba82af5251b7a2d6d28ecac1b3c5a8ad5a5e388353
+size 8209

138_128_e3_3e-5/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)