Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ 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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

config.json

@@ -0,0 +1,38 @@
+{
+  "architectures": [
+    "Qwen3MoeForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "decoder_sparse_step": 1,
+  "eos_token_id": 151643,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 48,
+  "mlp_only_layers": [],
+  "model_type": "qwen3_moe",
+  "moe_intermediate_size": 768,
+  "norm_topk_prob": true,
+  "num_attention_heads": 32,
+  "num_experts": 128,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 48,
+  "num_key_value_heads": 4,
+  "output_router_logits": false,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "router_aux_loss_coef": 0.001,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "max_new_tokens": 2048,
+  "transformers_version": "4.51.3"
+}

latest

@@ -0,0 +1 @@
+global_step314

model-00001-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98fb4fcaeb23558af36b1f8cf639c5068aaf4edfb62a8a0dd7341b571111d112
+size 4997184968

model-00002-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32922cba8b590d1d0c6b65d31924e36f1c2d21751f045626c9498a65d06ee4d2
+size 4997741608

model-00003-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb2ea3b306f552838fc59faa635ba390379aef8859888928d2afc5536419f976
+size 4997742208

model-00004-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08076872a45f7acfd4c30bae87792579523c34113a46588a1b8cc9cdb5f9ae1a
+size 4997743184

model-00005-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3db932bb8a0b7ff1c601dde28cf2e9aa74e158c38931dd3e2faf1a864b5e9b54
+size 4997743184

model-00006-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4db812bb9660dd1ebc3181f841d55d784227a2da7d73fb323c95b23d1286bd95
+size 4997743184

model-00007-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a1824296da6eeb28fd0889bc510cda06a29d6b9dd99b3391491a398e87b5501
+size 4997743184

model-00008-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94d865a40cff2a0a8ad8541aa97115fff8ec80658678356ca79b9f768a662bb4
+size 4997743184

model-00009-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2db6b06afcd9c87f99e75e1be59cfe767d78a7f14a109396457e0e782af21861
+size 4997743184

model-00010-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5e7a7c2b11d32c02713f17f5c23689c26bdd50a134ab2021ff1e5fb7b149be0
+size 4997743184

model-00011-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bd5e8b8f3fdba280b878761a8d7ec72185480dbe7276285b360250d15737908
+size 4997743184

model-00012-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0212f3794110b3129eceee9a0f5dcad983fdefe990c9044bb3634a20692174f6
+size 4997743184

model-00013-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65bc88c7f73e9178e17f6fd416f4954ada3cc98f07f1ec8cdfbb5f484e0e11d1
+size 1094220288

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:478b41e9f26d338fd8f896e08cad1adab7c423b61f1b45754113bc78d256a3f9
+size 16389

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce29a8767a7d907dd24987aa2c3e654d4317f3042fbc13b5b72cadb46d43311a
+size 16389

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61a48db011646b4e9a867bf12f4a233cad5dfbfe309686f8996c250196d3783a
+size 16389

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9562ee822472a4f01dcd6349ab3d1ef42a48915fe3b92e843a0c37db53c8421
+size 16389

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7d2767d83c3bf27f12db022b0632e2c4f8c164274ba75e380cf18f9d5f21819
+size 16389

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76816358d4e5db8149d60d55234db658d67a13c0c1ce05d7404cf7125a676a5c
+size 16389

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1562e7520c977d178183d641f70abcf3f57da2489938756cfbebf9b6e6c1a9fd
+size 16389

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6b6cabaed045c5398cd1b732f7ec48bd363f3b43cd24e0e70e641a42bd00c28
+size 16389

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8495648f53d8976e27a1c592199d8a3e553d3ae8d7dc9a78ce8b896d66bda62e
+size 1465

special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
+size 11422654

tokenizer_config.json

@@ -0,0 +1,241 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0].role == 'system' %}\n        {{- messages[0].content + '\\n\\n' }}\n    {%- endif %}\n    {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0].content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}\n        {%- set ns.multi_step_tool = false %}\n        {%- set ns.last_query_index = index %}\n    {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set content = message.content %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is defined and message.reasoning_content is not none %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in message.content %}\n                {%- set content = message.content.split('</think>')[-1].lstrip('\\n') %}\n                {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- if loop.index0 > ns.last_query_index %}\n            {%- if loop.last or (not loop.last and reasoning_content) %}\n                {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content.strip('\\n') + '\\n</think>\\n\\n' + content.lstrip('\\n') }}\n            {%- else %}\n                {{- '<|im_start|>' + message.role + '\\n' + content }}\n            {%- endif %}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if (loop.first and content) or (not loop.first) %}\n                    {{- '\\n' }}\n                {%- endif %}\n                {%- if tool_call.function %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {{- '<tool_call>\\n{\"name\": \"' }}\n                {{- tool_call.name }}\n                {{- '\", \"arguments\": ' }}\n                {%- if tool_call.arguments is string %}\n                    {{- tool_call.arguments }}\n                {%- else %}\n                    {{- tool_call.arguments | tojson }}\n                {%- endif %}\n                {{- '}\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is false %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

trainer_state.json

@@ -0,0 +1,2232 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6666666666666666,
+  "eval_steps": 500,
+  "global_step": 314,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0021231422505307855,
+      "grad_norm": 6.233692311689662,
+      "learning_rate": 0.0,
+      "loss": 1.3677,
+      "step": 1
+    },
+    {
+      "epoch": 0.004246284501061571,
+      "grad_norm": 6.036167846313234,
+      "learning_rate": 4.1666666666666667e-07,
+      "loss": 1.4092,
+      "step": 2
+    },
+    {
+      "epoch": 0.006369426751592357,
+      "grad_norm": 5.848179790677502,
+      "learning_rate": 8.333333333333333e-07,
+      "loss": 1.4005,
+      "step": 3
+    },
+    {
+      "epoch": 0.008492569002123142,
+      "grad_norm": 6.0378802369914935,
+      "learning_rate": 1.25e-06,
+      "loss": 1.4013,
+      "step": 4
+    },
+    {
+      "epoch": 0.010615711252653927,
+      "grad_norm": 5.595183474903656,
+      "learning_rate": 1.6666666666666667e-06,
+      "loss": 1.3827,
+      "step": 5
+    },
+    {
+      "epoch": 0.012738853503184714,
+      "grad_norm": 5.382553823715441,
+      "learning_rate": 2.0833333333333334e-06,
+      "loss": 1.4028,
+      "step": 6
+    },
+    {
+      "epoch": 0.014861995753715499,
+      "grad_norm": 5.302639986868994,
+      "learning_rate": 2.5e-06,
+      "loss": 1.2932,
+      "step": 7
+    },
+    {
+      "epoch": 0.016985138004246284,
+      "grad_norm": 4.567653411915884,
+      "learning_rate": 2.916666666666667e-06,
+      "loss": 1.3434,
+      "step": 8
+    },
+    {
+      "epoch": 0.01910828025477707,
+      "grad_norm": 4.288457445231689,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 1.3357,
+      "step": 9
+    },
+    {
+      "epoch": 0.021231422505307854,
+      "grad_norm": 4.049507973772927,
+      "learning_rate": 3.7500000000000005e-06,
+      "loss": 1.3412,
+      "step": 10
+    },
+    {
+      "epoch": 0.02335456475583864,
+      "grad_norm": 2.9217595476326017,
+      "learning_rate": 4.166666666666667e-06,
+      "loss": 1.2783,
+      "step": 11
+    },
+    {
+      "epoch": 0.025477707006369428,
+      "grad_norm": 3.005465009214924,
+      "learning_rate": 4.583333333333333e-06,
+      "loss": 1.2584,
+      "step": 12
+    },
+    {
+      "epoch": 0.027600849256900213,
+      "grad_norm": 2.793623616592523,
+      "learning_rate": 5e-06,
+      "loss": 1.3123,
+      "step": 13
+    },
+    {
+      "epoch": 0.029723991507430998,
+      "grad_norm": 2.090044767584013,
+      "learning_rate": 5.416666666666667e-06,
+      "loss": 1.2184,
+      "step": 14
+    },
+    {
+      "epoch": 0.03184713375796178,
+      "grad_norm": 2.6376790535616084,
+      "learning_rate": 5.833333333333334e-06,
+      "loss": 1.2336,
+      "step": 15
+    },
+    {
+      "epoch": 0.03397027600849257,
+      "grad_norm": 2.461503775389557,
+      "learning_rate": 6.25e-06,
+      "loss": 1.1937,
+      "step": 16
+    },
+    {
+      "epoch": 0.036093418259023353,
+      "grad_norm": 2.2913204705452395,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 1.158,
+      "step": 17
+    },
+    {
+      "epoch": 0.03821656050955414,
+      "grad_norm": 2.2503979169805084,
+      "learning_rate": 7.083333333333335e-06,
+      "loss": 1.1271,
+      "step": 18
+    },
+    {
+      "epoch": 0.040339702760084924,
+      "grad_norm": 2.466445150163946,
+      "learning_rate": 7.500000000000001e-06,
+      "loss": 1.1126,
+      "step": 19
+    },
+    {
+      "epoch": 0.04246284501061571,
+      "grad_norm": 2.247026245104246,
+      "learning_rate": 7.916666666666667e-06,
+      "loss": 1.0246,
+      "step": 20
+    },
+    {
+      "epoch": 0.044585987261146494,
+      "grad_norm": 2.3877387081949886,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 1.1415,
+      "step": 21
+    },
+    {
+      "epoch": 0.04670912951167728,
+      "grad_norm": 2.413953103563364,
+      "learning_rate": 8.750000000000001e-06,
+      "loss": 1.1117,
+      "step": 22
+    },
+    {
+      "epoch": 0.04883227176220807,
+      "grad_norm": 2.0737115581292063,
+      "learning_rate": 9.166666666666666e-06,
+      "loss": 1.0196,
+      "step": 23
+    },
+    {
+      "epoch": 0.050955414012738856,
+      "grad_norm": 1.4762727779050233,
+      "learning_rate": 9.583333333333335e-06,
+      "loss": 1.0301,
+      "step": 24
+    },
+    {
+      "epoch": 0.05307855626326964,
+      "grad_norm": 1.4318135959081477,
+      "learning_rate": 1e-05,
+      "loss": 1.0545,
+      "step": 25
+    },
+    {
+      "epoch": 0.055201698513800426,
+      "grad_norm": 4.215679400721786,
+      "learning_rate": 9.999876512522269e-06,
+      "loss": 1.046,
+      "step": 26
+    },
+    {
+      "epoch": 0.05732484076433121,
+      "grad_norm": 1.3411659207433844,
+      "learning_rate": 9.999506056188736e-06,
+      "loss": 1.089,
+      "step": 27
+    },
+    {
+      "epoch": 0.059447983014861996,
+      "grad_norm": 1.2807986819041255,
+      "learning_rate": 9.99888864929809e-06,
+      "loss": 1.0897,
+      "step": 28
+    },
+    {
+      "epoch": 0.06157112526539278,
+      "grad_norm": 1.1455939763312737,
+      "learning_rate": 9.99802432234714e-06,
+      "loss": 1.0569,
+      "step": 29
+    },
+    {
+      "epoch": 0.06369426751592357,
+      "grad_norm": 1.8215374393805115,
+      "learning_rate": 9.996913118029306e-06,
+      "loss": 1.0579,
+      "step": 30
+    },
+    {
+      "epoch": 0.06581740976645435,
+      "grad_norm": 1.0350186789586957,
+      "learning_rate": 9.995555091232516e-06,
+      "loss": 1.0399,
+      "step": 31
+    },
+    {
+      "epoch": 0.06794055201698514,
+      "grad_norm": 0.8818644928748289,
+      "learning_rate": 9.99395030903649e-06,
+      "loss": 1.0392,
+      "step": 32
+    },
+    {
+      "epoch": 0.07006369426751592,
+      "grad_norm": 0.9117340500480796,
+      "learning_rate": 9.992098850709434e-06,
+      "loss": 0.9961,
+      "step": 33
+    },
+    {
+      "epoch": 0.07218683651804671,
+      "grad_norm": 0.9087780060428899,
+      "learning_rate": 9.990000807704114e-06,
+      "loss": 0.9941,
+      "step": 34
+    },
+    {
+      "epoch": 0.07430997876857749,
+      "grad_norm": 0.8465013815457932,
+      "learning_rate": 9.987656283653344e-06,
+      "loss": 0.9976,
+      "step": 35
+    },
+    {
+      "epoch": 0.07643312101910828,
+      "grad_norm": 0.8845003722156789,
+      "learning_rate": 9.985065394364869e-06,
+      "loss": 1.0305,
+      "step": 36
+    },
+    {
+      "epoch": 0.07855626326963906,
+      "grad_norm": 0.9213919092982428,
+      "learning_rate": 9.982228267815644e-06,
+      "loss": 0.9694,
+      "step": 37
+    },
+    {
+      "epoch": 0.08067940552016985,
+      "grad_norm": 0.8874641991485241,
+      "learning_rate": 9.979145044145506e-06,
+      "loss": 0.9625,
+      "step": 38
+    },
+    {
+      "epoch": 0.08280254777070063,
+      "grad_norm": 0.8321483472211252,
+      "learning_rate": 9.975815875650265e-06,
+      "loss": 0.9474,
+      "step": 39
+    },
+    {
+      "epoch": 0.08492569002123142,
+      "grad_norm": 0.7808697647642835,
+      "learning_rate": 9.972240926774167e-06,
+      "loss": 0.998,
+      "step": 40
+    },
+    {
+      "epoch": 0.0870488322717622,
+      "grad_norm": 0.9040838208522329,
+      "learning_rate": 9.968420374101782e-06,
+      "loss": 0.9594,
+      "step": 41
+    },
+    {
+      "epoch": 0.08917197452229299,
+      "grad_norm": 0.7584425916438082,
+      "learning_rate": 9.964354406349272e-06,
+      "loss": 0.9628,
+      "step": 42
+    },
+    {
+      "epoch": 0.09129511677282377,
+      "grad_norm": 0.7620163299458729,
+      "learning_rate": 9.960043224355081e-06,
+      "loss": 0.9616,
+      "step": 43
+    },
+    {
+      "epoch": 0.09341825902335456,
+      "grad_norm": 0.8222242622111462,
+      "learning_rate": 9.955487041070003e-06,
+      "loss": 1.0289,
+      "step": 44
+    },
+    {
+      "epoch": 0.09554140127388536,
+      "grad_norm": 0.8421347235061095,
+      "learning_rate": 9.95068608154667e-06,
+      "loss": 1.0307,
+      "step": 45
+    },
+    {
+      "epoch": 0.09766454352441614,
+      "grad_norm": 0.9753656204578139,
+      "learning_rate": 9.945640582928438e-06,
+      "loss": 1.0164,
+      "step": 46
+    },
+    {
+      "epoch": 0.09978768577494693,
+      "grad_norm": 0.9292958552660628,
+      "learning_rate": 9.940350794437663e-06,
+      "loss": 0.9794,
+      "step": 47
+    },
+    {
+      "epoch": 0.10191082802547771,
+      "grad_norm": 1.1329023477241527,
+      "learning_rate": 9.934816977363404e-06,
+      "loss": 0.9825,
+      "step": 48
+    },
+    {
+      "epoch": 0.1040339702760085,
+      "grad_norm": 0.7480637668709176,
+      "learning_rate": 9.929039405048502e-06,
+      "loss": 0.9681,
+      "step": 49
+    },
+    {
+      "epoch": 0.10615711252653928,
+      "grad_norm": 0.8665970954982817,
+      "learning_rate": 9.923018362876093e-06,
+      "loss": 1.0051,
+      "step": 50
+    },
+    {
+      "epoch": 0.10828025477707007,
+      "grad_norm": 0.8205102895066251,
+      "learning_rate": 9.916754148255501e-06,
+      "loss": 0.9926,
+      "step": 51
+    },
+    {
+      "epoch": 0.11040339702760085,
+      "grad_norm": 0.7347897450905055,
+      "learning_rate": 9.91024707060755e-06,
+      "loss": 1.0003,
+      "step": 52
+    },
+    {
+      "epoch": 0.11252653927813164,
+      "grad_norm": 0.764879949996413,
+      "learning_rate": 9.903497451349286e-06,
+      "loss": 0.9292,
+      "step": 53
+    },
+    {
+      "epoch": 0.11464968152866242,
+      "grad_norm": 0.717706889231986,
+      "learning_rate": 9.896505623878088e-06,
+      "loss": 0.987,
+      "step": 54
+    },
+    {
+      "epoch": 0.11677282377919321,
+      "grad_norm": 0.7715458461663043,
+      "learning_rate": 9.889271933555214e-06,
+      "loss": 1.0261,
+      "step": 55
+    },
+    {
+      "epoch": 0.11889596602972399,
+      "grad_norm": 0.7521393701353877,
+      "learning_rate": 9.881796737688732e-06,
+      "loss": 1.0534,
+      "step": 56
+    },
+    {
+      "epoch": 0.12101910828025478,
+      "grad_norm": 0.7414115819906733,
+      "learning_rate": 9.874080405515874e-06,
+      "loss": 0.9913,
+      "step": 57
+    },
+    {
+      "epoch": 0.12314225053078556,
+      "grad_norm": 0.8232230218266363,
+      "learning_rate": 9.866123318184803e-06,
+      "loss": 0.9918,
+      "step": 58
+    },
+    {
+      "epoch": 0.12526539278131635,
+      "grad_norm": 0.8149896063868423,
+      "learning_rate": 9.857925868735774e-06,
+      "loss": 0.9868,
+      "step": 59
+    },
+    {
+      "epoch": 0.12738853503184713,
+      "grad_norm": 0.7942302058586811,
+      "learning_rate": 9.84948846208173e-06,
+      "loss": 0.98,
+      "step": 60
+    },
+    {
+      "epoch": 0.12951167728237792,
+      "grad_norm": 0.8138984220428053,
+      "learning_rate": 9.840811514988294e-06,
+      "loss": 0.9049,
+      "step": 61
+    },
+    {
+      "epoch": 0.1316348195329087,
+      "grad_norm": 0.7679169662702224,
+      "learning_rate": 9.831895456053197e-06,
+      "loss": 0.9677,
+      "step": 62
+    },
+    {
+      "epoch": 0.1337579617834395,
+      "grad_norm": 0.7874842581700767,
+      "learning_rate": 9.822740725685087e-06,
+      "loss": 0.9912,
+      "step": 63
+    },
+    {
+      "epoch": 0.13588110403397027,
+      "grad_norm": 0.7727018614004334,
+      "learning_rate": 9.81334777608179e-06,
+      "loss": 1.0131,
+      "step": 64
+    },
+    {
+      "epoch": 0.13800424628450106,
+      "grad_norm": 0.8942883107933416,
+      "learning_rate": 9.803717071207965e-06,
+      "loss": 1.022,
+      "step": 65
+    },
+    {
+      "epoch": 0.14012738853503184,
+      "grad_norm": 0.7488279786454368,
+      "learning_rate": 9.793849086772198e-06,
+      "loss": 0.9611,
+      "step": 66
+    },
+    {
+      "epoch": 0.14225053078556263,
+      "grad_norm": 0.7231313822633343,
+      "learning_rate": 9.783744310203492e-06,
+      "loss": 0.9648,
+      "step": 67
+    },
+    {
+      "epoch": 0.14437367303609341,
+      "grad_norm": 0.8009470186289236,
+      "learning_rate": 9.77340324062719e-06,
+      "loss": 0.9616,
+      "step": 68
+    },
+    {
+      "epoch": 0.1464968152866242,
+      "grad_norm": 0.694840815258157,
+      "learning_rate": 9.76282638884034e-06,
+      "loss": 0.9393,
+      "step": 69
+    },
+    {
+      "epoch": 0.14861995753715498,
+      "grad_norm": 0.807816808271673,
+      "learning_rate": 9.752014277286433e-06,
+      "loss": 1.0505,
+      "step": 70
+    },
+    {
+      "epoch": 0.15074309978768577,
+      "grad_norm": 0.8317043459433457,
+      "learning_rate": 9.740967440029628e-06,
+      "loss": 0.9914,
+      "step": 71
+    },
+    {
+      "epoch": 0.15286624203821655,
+      "grad_norm": 0.6790810524359902,
+      "learning_rate": 9.729686422728353e-06,
+      "loss": 1.012,
+      "step": 72
+    },
+    {
+      "epoch": 0.15498938428874734,
+      "grad_norm": 0.7310970501009117,
+      "learning_rate": 9.718171782608355e-06,
+      "loss": 0.959,
+      "step": 73
+    },
+    {
+      "epoch": 0.15711252653927812,
+      "grad_norm": 0.7675118911080142,
+      "learning_rate": 9.706424088435183e-06,
+      "loss": 0.9373,
+      "step": 74
+    },
+    {
+      "epoch": 0.1592356687898089,
+      "grad_norm": 0.6926951658050571,
+      "learning_rate": 9.694443920486083e-06,
+      "loss": 0.9801,
+      "step": 75
+    },
+    {
+      "epoch": 0.1613588110403397,
+      "grad_norm": 0.7346879776016684,
+      "learning_rate": 9.682231870521347e-06,
+      "loss": 0.8983,
+      "step": 76
+    },
+    {
+      "epoch": 0.16348195329087048,
+      "grad_norm": 0.6862950789085892,
+      "learning_rate": 9.669788541755072e-06,
+      "loss": 0.9532,
+      "step": 77
+    },
+    {
+      "epoch": 0.16560509554140126,
+      "grad_norm": 0.7381966173799152,
+      "learning_rate": 9.657114548825372e-06,
+      "loss": 0.9686,
+      "step": 78
+    },
+    {
+      "epoch": 0.16772823779193205,
+      "grad_norm": 0.643557401853306,
+      "learning_rate": 9.644210517764014e-06,
+      "loss": 0.9668,
+      "step": 79
+    },
+    {
+      "epoch": 0.16985138004246284,
+      "grad_norm": 0.7765772003310986,
+      "learning_rate": 9.631077085965501e-06,
+      "loss": 1.0638,
+      "step": 80
+    },
+    {
+      "epoch": 0.17197452229299362,
+      "grad_norm": 0.8024757246296556,
+      "learning_rate": 9.617714902155576e-06,
+      "loss": 0.9911,
+      "step": 81
+    },
+    {
+      "epoch": 0.1740976645435244,
+      "grad_norm": 0.7800601554968952,
+      "learning_rate": 9.60412462635919e-06,
+      "loss": 1.0459,
+      "step": 82
+    },
+    {
+      "epoch": 0.1762208067940552,
+      "grad_norm": 0.7431393505154594,
+      "learning_rate": 9.590306929867896e-06,
+      "loss": 0.9687,
+      "step": 83
+    },
+    {
+      "epoch": 0.17834394904458598,
+      "grad_norm": 0.6824175459850116,
+      "learning_rate": 9.576262495206689e-06,
+      "loss": 0.9994,
+      "step": 84
+    },
+    {
+      "epoch": 0.18046709129511676,
+      "grad_norm": 0.7916336363488418,
+      "learning_rate": 9.561992016100293e-06,
+      "loss": 0.9783,
+      "step": 85
+    },
+    {
+      "epoch": 0.18259023354564755,
+      "grad_norm": 0.6976948182235398,
+      "learning_rate": 9.547496197438896e-06,
+      "loss": 1.0076,
+      "step": 86
+    },
+    {
+      "epoch": 0.18471337579617833,
+      "grad_norm": 0.7093988344101032,
+      "learning_rate": 9.532775755243334e-06,
+      "loss": 1.0075,
+      "step": 87
+    },
+    {
+      "epoch": 0.18683651804670912,
+      "grad_norm": 0.752335224864414,
+      "learning_rate": 9.517831416629717e-06,
+      "loss": 0.9428,
+      "step": 88
+    },
+    {
+      "epoch": 0.18895966029723993,
+      "grad_norm": 0.7535135712750508,
+      "learning_rate": 9.502663919773516e-06,
+      "loss": 0.9228,
+      "step": 89
+    },
+    {
+      "epoch": 0.1910828025477707,
+      "grad_norm": 0.6836684618511217,
+      "learning_rate": 9.487274013873104e-06,
+      "loss": 0.9915,
+      "step": 90
+    },
+    {
+      "epoch": 0.1932059447983015,
+      "grad_norm": 0.7168371002333184,
+      "learning_rate": 9.471662459112747e-06,
+      "loss": 0.9146,
+      "step": 91
+    },
+    {
+      "epoch": 0.19532908704883228,
+      "grad_norm": 0.7590517179947264,
+      "learning_rate": 9.455830026625053e-06,
+      "loss": 0.9537,
+      "step": 92
+    },
+    {
+      "epoch": 0.19745222929936307,
+      "grad_norm": 0.783311626765522,
+      "learning_rate": 9.439777498452883e-06,
+      "loss": 0.9673,
+      "step": 93
+    },
+    {
+      "epoch": 0.19957537154989385,
+      "grad_norm": 0.7583109403403397,
+      "learning_rate": 9.423505667510724e-06,
+      "loss": 0.9488,
+      "step": 94
+    },
+    {
+      "epoch": 0.20169851380042464,
+      "grad_norm": 0.6665250800247178,
+      "learning_rate": 9.40701533754552e-06,
+      "loss": 1.0014,
+      "step": 95
+    },
+    {
+      "epoch": 0.20382165605095542,
+      "grad_norm": 0.7534757180486034,
+      "learning_rate": 9.390307323096972e-06,
+      "loss": 0.9486,
+      "step": 96
+    },
+    {
+      "epoch": 0.2059447983014862,
+      "grad_norm": 0.7633774921562732,
+      "learning_rate": 9.373382449457305e-06,
+      "loss": 0.9486,
+      "step": 97
+    },
+    {
+      "epoch": 0.208067940552017,
+      "grad_norm": 0.7546302945571557,
+      "learning_rate": 9.356241552630503e-06,
+      "loss": 0.9325,
+      "step": 98
+    },
+    {
+      "epoch": 0.21019108280254778,
+      "grad_norm": 0.7374682216888635,
+      "learning_rate": 9.338885479291012e-06,
+      "loss": 1.0019,
+      "step": 99
+    },
+    {
+      "epoch": 0.21231422505307856,
+      "grad_norm": 0.7502571413886471,
+      "learning_rate": 9.321315086741916e-06,
+      "loss": 1.0028,
+      "step": 100
+    },
+    {
+      "epoch": 0.21443736730360935,
+      "grad_norm": 0.7215038325133775,
+      "learning_rate": 9.303531242872606e-06,
+      "loss": 1.038,
+      "step": 101
+    },
+    {
+      "epoch": 0.21656050955414013,
+      "grad_norm": 0.7766452530718216,
+      "learning_rate": 9.285534826115884e-06,
+      "loss": 0.9567,
+      "step": 102
+    },
+    {
+      "epoch": 0.21868365180467092,
+      "grad_norm": 0.7660906470554663,
+      "learning_rate": 9.2673267254046e-06,
+      "loss": 0.9041,
+      "step": 103
+    },
+    {
+      "epoch": 0.2208067940552017,
+      "grad_norm": 0.6968152389503037,
+      "learning_rate": 9.248907840127726e-06,
+      "loss": 0.9682,
+      "step": 104
+    },
+    {
+      "epoch": 0.2229299363057325,
+      "grad_norm": 0.6843474144819354,
+      "learning_rate": 9.230279080085933e-06,
+      "loss": 0.9289,
+      "step": 105
+    },
+    {
+      "epoch": 0.22505307855626328,
+      "grad_norm": 0.7512996746200725,
+      "learning_rate": 9.211441365446661e-06,
+      "loss": 0.9318,
+      "step": 106
+    },
+    {
+      "epoch": 0.22717622080679406,
+      "grad_norm": 0.7245746774218922,
+      "learning_rate": 9.192395626698656e-06,
+      "loss": 0.9745,
+      "step": 107
+    },
+    {
+      "epoch": 0.22929936305732485,
+      "grad_norm": 0.7472828991840396,
+      "learning_rate": 9.173142804606012e-06,
+      "loss": 0.9417,
+      "step": 108
+    },
+    {
+      "epoch": 0.23142250530785563,
+      "grad_norm": 0.7433451614104158,
+      "learning_rate": 9.153683850161706e-06,
+      "loss": 0.9888,
+      "step": 109
+    },
+    {
+      "epoch": 0.23354564755838642,
+      "grad_norm": 0.6820085118229582,
+      "learning_rate": 9.13401972454062e-06,
+      "loss": 1.0062,
+      "step": 110
+    },
+    {
+      "epoch": 0.2356687898089172,
+      "grad_norm": 0.6597210654799798,
+      "learning_rate": 9.114151399052064e-06,
+      "loss": 0.9498,
+      "step": 111
+    },
+    {
+      "epoch": 0.23779193205944799,
+      "grad_norm": 0.6964084111810886,
+      "learning_rate": 9.094079855091797e-06,
+      "loss": 0.9684,
+      "step": 112
+    },
+    {
+      "epoch": 0.23991507430997877,
+      "grad_norm": 0.7723142466634979,
+      "learning_rate": 9.073806084093556e-06,
+      "loss": 0.9405,
+      "step": 113
+    },
+    {
+      "epoch": 0.24203821656050956,
+      "grad_norm": 0.7911498715522838,
+      "learning_rate": 9.053331087480075e-06,
+      "loss": 0.9474,
+      "step": 114
+    },
+    {
+      "epoch": 0.24416135881104034,
+      "grad_norm": 0.7154015242350461,
+      "learning_rate": 9.032655876613636e-06,
+      "loss": 0.9994,
+      "step": 115
+    },
+    {
+      "epoch": 0.24628450106157113,
+      "grad_norm": 0.8131533865526955,
+      "learning_rate": 9.01178147274609e-06,
+      "loss": 0.9674,
+      "step": 116
+    },
+    {
+      "epoch": 0.2484076433121019,
+      "grad_norm": 0.762660871457576,
+      "learning_rate": 8.990708906968431e-06,
+      "loss": 0.9712,
+      "step": 117
+    },
+    {
+      "epoch": 0.2505307855626327,
+      "grad_norm": 0.749279770619595,
+      "learning_rate": 8.969439220159861e-06,
+      "loss": 0.9718,
+      "step": 118
+    },
+    {
+      "epoch": 0.2526539278131635,
+      "grad_norm": 0.7993155759440779,
+      "learning_rate": 8.947973462936366e-06,
+      "loss": 1.002,
+      "step": 119
+    },
+    {
+      "epoch": 0.25477707006369427,
+      "grad_norm": 0.7525506488087893,
+      "learning_rate": 8.926312695598837e-06,
+      "loss": 1.0678,
+      "step": 120
+    },
+    {
+      "epoch": 0.25690021231422505,
+      "grad_norm": 0.7942187041189313,
+      "learning_rate": 8.904457988080682e-06,
+      "loss": 0.9531,
+      "step": 121
+    },
+    {
+      "epoch": 0.25902335456475584,
+      "grad_norm": 0.725938463845709,
+      "learning_rate": 8.882410419894983e-06,
+      "loss": 0.9536,
+      "step": 122
+    },
+    {
+      "epoch": 0.2611464968152866,
+      "grad_norm": 0.6926002729927859,
+      "learning_rate": 8.860171080081174e-06,
+      "loss": 1.0066,
+      "step": 123
+    },
+    {
+      "epoch": 0.2632696390658174,
+      "grad_norm": 0.7073710959096725,
+      "learning_rate": 8.837741067151251e-06,
+      "loss": 0.9731,
+      "step": 124
+    },
+    {
+      "epoch": 0.2653927813163482,
+      "grad_norm": 1.229783488810088,
+      "learning_rate": 8.8151214890355e-06,
+      "loss": 0.9316,
+      "step": 125
+    },
+    {
+      "epoch": 0.267515923566879,
+      "grad_norm": 0.673125965324655,
+      "learning_rate": 8.792313463027777e-06,
+      "loss": 1.0252,
+      "step": 126
+    },
+    {
+      "epoch": 0.26963906581740976,
+      "grad_norm": 0.6930092507451997,
+      "learning_rate": 8.76931811573033e-06,
+      "loss": 0.9374,
+      "step": 127
+    },
+    {
+      "epoch": 0.27176220806794055,
+      "grad_norm": 0.68412243336128,
+      "learning_rate": 8.74613658299813e-06,
+      "loss": 0.8807,
+      "step": 128
+    },
+    {
+      "epoch": 0.27388535031847133,
+      "grad_norm": 0.7081201802451947,
+      "learning_rate": 8.72277000988278e-06,
+      "loss": 0.9641,
+      "step": 129
+    },
+    {
+      "epoch": 0.2760084925690021,
+      "grad_norm": 0.6854965625822327,
+      "learning_rate": 8.699219550575954e-06,
+      "loss": 0.9761,
+      "step": 130
+    },
+    {
+      "epoch": 0.2781316348195329,
+      "grad_norm": 0.6538579090164267,
+      "learning_rate": 8.675486368352376e-06,
+      "loss": 0.9509,
+      "step": 131
+    },
+    {
+      "epoch": 0.2802547770700637,
+      "grad_norm": 0.6798135259731691,
+      "learning_rate": 8.651571635512372e-06,
+      "loss": 0.9736,
+      "step": 132
+    },
+    {
+      "epoch": 0.2823779193205945,
+      "grad_norm": 0.6711453015182969,
+      "learning_rate": 8.627476533323957e-06,
+      "loss": 0.9235,
+      "step": 133
+    },
+    {
+      "epoch": 0.28450106157112526,
+      "grad_norm": 0.7583804715274035,
+      "learning_rate": 8.603202251964492e-06,
+      "loss": 0.9981,
+      "step": 134
+    },
+    {
+      "epoch": 0.28662420382165604,
+      "grad_norm": 0.7176651957782098,
+      "learning_rate": 8.578749990461884e-06,
+      "loss": 1.0032,
+      "step": 135
+    },
+    {
+      "epoch": 0.28874734607218683,
+      "grad_norm": 0.7211103037306501,
+      "learning_rate": 8.554120956635375e-06,
+      "loss": 0.9909,
+      "step": 136
+    },
+    {
+      "epoch": 0.2908704883227176,
+      "grad_norm": 0.688485015985765,
+      "learning_rate": 8.52931636703587e-06,
+      "loss": 0.9276,
+      "step": 137
+    },
+    {
+      "epoch": 0.2929936305732484,
+      "grad_norm": 0.6465279506859974,
+      "learning_rate": 8.504337446885854e-06,
+      "loss": 0.9398,
+      "step": 138
+    },
+    {
+      "epoch": 0.2951167728237792,
+      "grad_norm": 0.6867853315980083,
+      "learning_rate": 8.47918543001886e-06,
+      "loss": 0.9279,
+      "step": 139
+    },
+    {
+      "epoch": 0.29723991507430997,
+      "grad_norm": 0.7122704321655888,
+      "learning_rate": 8.453861558818542e-06,
+      "loss": 0.9884,
+      "step": 140
+    },
+    {
+      "epoch": 0.29936305732484075,
+      "grad_norm": 0.7229336013729085,
+      "learning_rate": 8.428367084157292e-06,
+      "loss": 0.9288,
+      "step": 141
+    },
+    {
+      "epoch": 0.30148619957537154,
+      "grad_norm": 0.6611042572044565,
+      "learning_rate": 8.402703265334455e-06,
+      "loss": 1.0012,
+      "step": 142
+    },
+    {
+      "epoch": 0.3036093418259023,
+      "grad_norm": 0.6705484140078379,
+      "learning_rate": 8.376871370014139e-06,
+      "loss": 1.0219,
+      "step": 143
+    },
+    {
+      "epoch": 0.3057324840764331,
+      "grad_norm": 0.7118638939797448,
+      "learning_rate": 8.350872674162578e-06,
+      "loss": 0.8994,
+      "step": 144
+    },
+    {
+      "epoch": 0.3078556263269639,
+      "grad_norm": 0.7284734778507161,
+      "learning_rate": 8.324708461985124e-06,
+      "loss": 0.9674,
+      "step": 145
+    },
+    {
+      "epoch": 0.3099787685774947,
+      "grad_norm": 0.723907696270781,
+      "learning_rate": 8.298380025862805e-06,
+      "loss": 0.9132,
+      "step": 146
+    },
+    {
+      "epoch": 0.31210191082802546,
+      "grad_norm": 0.7559028348047754,
+      "learning_rate": 8.271888666288488e-06,
+      "loss": 0.94,
+      "step": 147
+    },
+    {
+      "epoch": 0.31422505307855625,
+      "grad_norm": 0.6723280638514304,
+      "learning_rate": 8.245235691802644e-06,
+      "loss": 1.0277,
+      "step": 148
+    },
+    {
+      "epoch": 0.31634819532908703,
+      "grad_norm": 0.7371820881632951,
+      "learning_rate": 8.218422418928709e-06,
+      "loss": 1.0206,
+      "step": 149
+    },
+    {
+      "epoch": 0.3184713375796178,
+      "grad_norm": 0.6827845610709571,
+      "learning_rate": 8.191450172108058e-06,
+      "loss": 0.9469,
+      "step": 150
+    },
+    {
+      "epoch": 0.3205944798301486,
+      "grad_norm": 0.7055002715463514,
+      "learning_rate": 8.164320283634585e-06,
+      "loss": 0.9841,
+      "step": 151
+    },
+    {
+      "epoch": 0.3227176220806794,
+      "grad_norm": 0.6889063122118256,
+      "learning_rate": 8.137034093588885e-06,
+      "loss": 0.984,
+      "step": 152
+    },
+    {
+      "epoch": 0.3248407643312102,
+      "grad_norm": 0.6966955432660643,
+      "learning_rate": 8.109592949772076e-06,
+      "loss": 1.013,
+      "step": 153
+    },
+    {
+      "epoch": 0.32696390658174096,
+      "grad_norm": 0.6986846471550263,
+      "learning_rate": 8.081998207639212e-06,
+      "loss": 1.0059,
+      "step": 154
+    },
+    {
+      "epoch": 0.32908704883227174,
+      "grad_norm": 0.626142059105963,
+      "learning_rate": 8.054251230232333e-06,
+      "loss": 0.9991,
+      "step": 155
+    },
+    {
+      "epoch": 0.33121019108280253,
+      "grad_norm": 0.7025129741121969,
+      "learning_rate": 8.026353388113142e-06,
+      "loss": 0.9797,
+      "step": 156
+    },
+    {
+      "epoch": 0.3333333333333333,
+      "grad_norm": 0.6696025000651015,
+      "learning_rate": 7.998306059295302e-06,
+      "loss": 1.0307,
+      "step": 157
+    },
+    {
+      "epoch": 0.3354564755838641,
+      "grad_norm": 0.6273584551839821,
+      "learning_rate": 7.97011062917637e-06,
+      "loss": 0.9362,
+      "step": 158
+    },
+    {
+      "epoch": 0.3375796178343949,
+      "grad_norm": 0.702578261973236,
+      "learning_rate": 7.941768490469368e-06,
+      "loss": 1.0164,
+      "step": 159
+    },
+    {
+      "epoch": 0.33970276008492567,
+      "grad_norm": 0.7406287649359133,
+      "learning_rate": 7.913281043133978e-06,
+      "loss": 0.9494,
+      "step": 160
+    },
+    {
+      "epoch": 0.34182590233545646,
+      "grad_norm": 0.6752561053160435,
+      "learning_rate": 7.884649694307413e-06,
+      "loss": 0.916,
+      "step": 161
+    },
+    {
+      "epoch": 0.34394904458598724,
+      "grad_norm": 0.6827707894296586,
+      "learning_rate": 7.855875858234894e-06,
+      "loss": 0.957,
+      "step": 162
+    },
+    {
+      "epoch": 0.346072186836518,
+      "grad_norm": 0.721736590005719,
+      "learning_rate": 7.826960956199796e-06,
+      "loss": 0.9642,
+      "step": 163
+    },
+    {
+      "epoch": 0.3481953290870488,
+      "grad_norm": 0.6861052725661058,
+      "learning_rate": 7.797906416453445e-06,
+      "loss": 1.0046,
+      "step": 164
+    },
+    {
+      "epoch": 0.3503184713375796,
+      "grad_norm": 0.6571115602420129,
+      "learning_rate": 7.768713674144578e-06,
+      "loss": 1.0109,
+      "step": 165
+    },
+    {
+      "epoch": 0.3524416135881104,
+      "grad_norm": 0.6826123583598673,
+      "learning_rate": 7.739384171248436e-06,
+      "loss": 0.9203,
+      "step": 166
+    },
+    {
+      "epoch": 0.35456475583864117,
+      "grad_norm": 0.7528479464049002,
+      "learning_rate": 7.709919356495555e-06,
+      "loss": 0.954,
+      "step": 167
+    },
+    {
+      "epoch": 0.35668789808917195,
+      "grad_norm": 0.6026943812612681,
+      "learning_rate": 7.6803206853002e-06,
+      "loss": 1.0024,
+      "step": 168
+    },
+    {
+      "epoch": 0.35881104033970274,
+      "grad_norm": 0.7023232885151509,
+      "learning_rate": 7.650589619688468e-06,
+      "loss": 0.9751,
+      "step": 169
+    },
+    {
+      "epoch": 0.3609341825902335,
+      "grad_norm": 0.6635390470590239,
+      "learning_rate": 7.620727628226081e-06,
+      "loss": 0.908,
+      "step": 170
+    },
+    {
+      "epoch": 0.3630573248407643,
+      "grad_norm": 0.6202385073922436,
+      "learning_rate": 7.590736185945843e-06,
+      "loss": 0.9455,
+      "step": 171
+    },
+    {
+      "epoch": 0.3651804670912951,
+      "grad_norm": 0.7074191000317777,
+      "learning_rate": 7.560616774274775e-06,
+      "loss": 0.9342,
+      "step": 172
+    },
+    {
+      "epoch": 0.3673036093418259,
+      "grad_norm": 0.6912450203775482,
+      "learning_rate": 7.5303708809609514e-06,
+      "loss": 0.9303,
+      "step": 173
+    },
+    {
+      "epoch": 0.36942675159235666,
+      "grad_norm": 0.6716646015086942,
+      "learning_rate": 7.500000000000001e-06,
+      "loss": 0.9717,
+      "step": 174
+    },
+    {
+      "epoch": 0.37154989384288745,
+      "grad_norm": 0.6478930551539204,
+      "learning_rate": 7.469505631561318e-06,
+      "loss": 0.9443,
+      "step": 175
+    },
+    {
+      "epoch": 0.37367303609341823,
+      "grad_norm": 0.7036455495923415,
+      "learning_rate": 7.4388892819139625e-06,
+      "loss": 0.981,
+      "step": 176
+    },
+    {
+      "epoch": 0.37579617834394907,
+      "grad_norm": 0.6998836705652469,
+      "learning_rate": 7.408152463352249e-06,
+      "loss": 0.9392,
+      "step": 177
+    },
+    {
+      "epoch": 0.37791932059447986,
+      "grad_norm": 0.7177208792525537,
+      "learning_rate": 7.3772966941210585e-06,
+      "loss": 0.9794,
+      "step": 178
+    },
+    {
+      "epoch": 0.38004246284501064,
+      "grad_norm": 0.7473525068625702,
+      "learning_rate": 7.346323498340839e-06,
+      "loss": 0.9195,
+      "step": 179
+    },
+    {
+      "epoch": 0.3821656050955414,
+      "grad_norm": 0.7227554431825605,
+      "learning_rate": 7.3152344059323165e-06,
+      "loss": 1.0082,
+      "step": 180
+    },
+    {
+      "epoch": 0.3842887473460722,
+      "grad_norm": 0.7080895952572748,
+      "learning_rate": 7.284030952540937e-06,
+      "loss": 0.9767,
+      "step": 181
+    },
+    {
+      "epoch": 0.386411889596603,
+      "grad_norm": 0.6629730047476243,
+      "learning_rate": 7.252714679461001e-06,
+      "loss": 0.994,
+      "step": 182
+    },
+    {
+      "epoch": 0.3885350318471338,
+      "grad_norm": 0.7045119821135271,
+      "learning_rate": 7.221287133559537e-06,
+      "loss": 0.9305,
+      "step": 183
+    },
+    {
+      "epoch": 0.39065817409766457,
+      "grad_norm": 0.6763415144165187,
+      "learning_rate": 7.189749867199899e-06,
+      "loss": 0.9091,
+      "step": 184
+    },
+    {
+      "epoch": 0.39278131634819535,
+      "grad_norm": 0.7251593181916779,
+      "learning_rate": 7.1581044381650735e-06,
+      "loss": 0.9801,
+      "step": 185
+    },
+    {
+      "epoch": 0.39490445859872614,
+      "grad_norm": 0.7318304915524612,
+      "learning_rate": 7.126352409580749e-06,
+      "loss": 1.0148,
+      "step": 186
+    },
+    {
+      "epoch": 0.3970276008492569,
+      "grad_norm": 0.636231931630425,
+      "learning_rate": 7.094495349838093e-06,
+      "loss": 1.0049,
+      "step": 187
+    },
+    {
+      "epoch": 0.3991507430997877,
+      "grad_norm": 0.7158842207998385,
+      "learning_rate": 7.062534832516288e-06,
+      "loss": 1.0129,
+      "step": 188
+    },
+    {
+      "epoch": 0.4012738853503185,
+      "grad_norm": 0.7129569595828432,
+      "learning_rate": 7.0304724363048025e-06,
+      "loss": 0.9263,
+      "step": 189
+    },
+    {
+      "epoch": 0.4033970276008493,
+      "grad_norm": 0.6864707114810514,
+      "learning_rate": 6.998309744925411e-06,
+      "loss": 0.8735,
+      "step": 190
+    },
+    {
+      "epoch": 0.40552016985138006,
+      "grad_norm": 0.6698406211001818,
+      "learning_rate": 6.9660483470539704e-06,
+      "loss": 0.9367,
+      "step": 191
+    },
+    {
+      "epoch": 0.40764331210191085,
+      "grad_norm": 0.7267969373273135,
+      "learning_rate": 6.933689836241939e-06,
+      "loss": 0.9406,
+      "step": 192
+    },
+    {
+      "epoch": 0.40976645435244163,
+      "grad_norm": 0.651813445608617,
+      "learning_rate": 6.901235810837668e-06,
+      "loss": 0.9736,
+      "step": 193
+    },
+    {
+      "epoch": 0.4118895966029724,
+      "grad_norm": 0.6683944283747483,
+      "learning_rate": 6.868687873907458e-06,
+      "loss": 0.9579,
+      "step": 194
+    },
+    {
+      "epoch": 0.4140127388535032,
+      "grad_norm": 0.6506700841660094,
+      "learning_rate": 6.836047633156361e-06,
+      "loss": 0.9174,
+      "step": 195
+    },
+    {
+      "epoch": 0.416135881104034,
+      "grad_norm": 0.6616818716028071,
+      "learning_rate": 6.8033167008487784e-06,
+      "loss": 0.9087,
+      "step": 196
+    },
+    {
+      "epoch": 0.4182590233545648,
+      "grad_norm": 0.6605052606554319,
+      "learning_rate": 6.77049669372882e-06,
+      "loss": 0.9679,
+      "step": 197
+    },
+    {
+      "epoch": 0.42038216560509556,
+      "grad_norm": 0.7082727940149344,
+      "learning_rate": 6.737589232940445e-06,
+      "loss": 0.9985,
+      "step": 198
+    },
+    {
+      "epoch": 0.42250530785562634,
+      "grad_norm": 0.6623600494563191,
+      "learning_rate": 6.704595943947385e-06,
+      "loss": 0.9321,
+      "step": 199
+    },
+    {
+      "epoch": 0.42462845010615713,
+      "grad_norm": 0.6623978762054843,
+      "learning_rate": 6.671518456452859e-06,
+      "loss": 0.9501,
+      "step": 200
+    },
+    {
+      "epoch": 0.4267515923566879,
+      "grad_norm": 0.6742018229334826,
+      "learning_rate": 6.638358404319064e-06,
+      "loss": 0.9684,
+      "step": 201
+    },
+    {
+      "epoch": 0.4288747346072187,
+      "grad_norm": 0.6502309799062719,
+      "learning_rate": 6.605117425486483e-06,
+      "loss": 0.9631,
+      "step": 202
+    },
+    {
+      "epoch": 0.4309978768577495,
+      "grad_norm": 0.6259967571338436,
+      "learning_rate": 6.571797161892965e-06,
+      "loss": 0.9195,
+      "step": 203
+    },
+    {
+      "epoch": 0.43312101910828027,
+      "grad_norm": 0.6800317584945854,
+      "learning_rate": 6.538399259392637e-06,
+      "loss": 0.9516,
+      "step": 204
+    },
+    {
+      "epoch": 0.43524416135881105,
+      "grad_norm": 0.6261983185852475,
+      "learning_rate": 6.504925367674595e-06,
+      "loss": 1.0328,
+      "step": 205
+    },
+    {
+      "epoch": 0.43736730360934184,
+      "grad_norm": 0.6449729939238322,
+      "learning_rate": 6.471377140181419e-06,
+      "loss": 0.9397,
+      "step": 206
+    },
+    {
+      "epoch": 0.4394904458598726,
+      "grad_norm": 0.706583045696564,
+      "learning_rate": 6.437756234027512e-06,
+      "loss": 0.9474,
+      "step": 207
+    },
+    {
+      "epoch": 0.4416135881104034,
+      "grad_norm": 0.6919864898373091,
+      "learning_rate": 6.40406430991723e-06,
+      "loss": 0.91,
+      "step": 208
+    },
+    {
+      "epoch": 0.4437367303609342,
+      "grad_norm": 0.6613982934744788,
+      "learning_rate": 6.370303032062869e-06,
+      "loss": 0.9095,
+      "step": 209
+    },
+    {
+      "epoch": 0.445859872611465,
+      "grad_norm": 0.6736801782820888,
+      "learning_rate": 6.336474068102444e-06,
+      "loss": 0.9731,
+      "step": 210
+    },
+    {
+      "epoch": 0.44798301486199577,
+      "grad_norm": 0.7089874077887519,
+      "learning_rate": 6.302579089017328e-06,
+      "loss": 0.9463,
+      "step": 211
+    },
+    {
+      "epoch": 0.45010615711252655,
+      "grad_norm": 0.7181675289310537,
+      "learning_rate": 6.268619769049713e-06,
+      "loss": 0.922,
+      "step": 212
+    },
+    {
+      "epoch": 0.45222929936305734,
+      "grad_norm": 0.7018430577826239,
+      "learning_rate": 6.234597785619906e-06,
+      "loss": 0.9219,
+      "step": 213
+    },
+    {
+      "epoch": 0.4543524416135881,
+      "grad_norm": 0.6853546290531675,
+      "learning_rate": 6.200514819243476e-06,
+      "loss": 0.907,
+      "step": 214
+    },
+    {
+      "epoch": 0.4564755838641189,
+      "grad_norm": 0.686355248670492,
+      "learning_rate": 6.166372553448241e-06,
+      "loss": 0.9159,
+      "step": 215
+    },
+    {
+      "epoch": 0.4585987261146497,
+      "grad_norm": 0.7339054174560321,
+      "learning_rate": 6.132172674691119e-06,
+      "loss": 0.9645,
+      "step": 216
+    },
+    {
+      "epoch": 0.4607218683651805,
+      "grad_norm": 0.6572741843944705,
+      "learning_rate": 6.097916872274815e-06,
+      "loss": 1.005,
+      "step": 217
+    },
+    {
+      "epoch": 0.46284501061571126,
+      "grad_norm": 0.6385617975237159,
+      "learning_rate": 6.063606838264384e-06,
+      "loss": 0.9395,
+      "step": 218
+    },
+    {
+      "epoch": 0.46496815286624205,
+      "grad_norm": 0.6298680692637662,
+      "learning_rate": 6.029244267403652e-06,
+      "loss": 0.921,
+      "step": 219
+    },
+    {
+      "epoch": 0.46709129511677283,
+      "grad_norm": 0.6537484240328865,
+      "learning_rate": 5.9948308570315e-06,
+      "loss": 0.9418,
+      "step": 220
+    },
+    {
+      "epoch": 0.4692144373673036,
+      "grad_norm": 0.6513186789442099,
+      "learning_rate": 5.960368306998023e-06,
+      "loss": 1.0093,
+      "step": 221
+    },
+    {
+      "epoch": 0.4713375796178344,
+      "grad_norm": 0.6198436120565162,
+      "learning_rate": 5.92585831958058e-06,
+      "loss": 0.8843,
+      "step": 222
+    },
+    {
+      "epoch": 0.4734607218683652,
+      "grad_norm": 0.6949192129540798,
+      "learning_rate": 5.891302599399686e-06,
+      "loss": 0.9173,
+      "step": 223
+    },
+    {
+      "epoch": 0.47558386411889597,
+      "grad_norm": 0.661048566711158,
+      "learning_rate": 5.856702853334833e-06,
+      "loss": 0.993,
+      "step": 224
+    },
+    {
+      "epoch": 0.47770700636942676,
+      "grad_norm": 0.6928421971972144,
+      "learning_rate": 5.8220607904401725e-06,
+      "loss": 0.9237,
+      "step": 225
+    },
+    {
+      "epoch": 0.47983014861995754,
+      "grad_norm": 0.6766405995371992,
+      "learning_rate": 5.78737812186009e-06,
+      "loss": 0.9212,
+      "step": 226
+    },
+    {
+      "epoch": 0.4819532908704883,
+      "grad_norm": 0.6387732380032056,
+      "learning_rate": 5.752656560744692e-06,
+      "loss": 0.9563,
+      "step": 227
+    },
+    {
+      "epoch": 0.4840764331210191,
+      "grad_norm": 0.6801696449031966,
+      "learning_rate": 5.717897822165179e-06,
+      "loss": 0.9556,
+      "step": 228
+    },
+    {
+      "epoch": 0.4861995753715499,
+      "grad_norm": 0.7117883554839965,
+      "learning_rate": 5.6831036230291345e-06,
+      "loss": 0.9756,
+      "step": 229
+    },
+    {
+      "epoch": 0.4883227176220807,
+      "grad_norm": 0.7055614143035667,
+      "learning_rate": 5.648275681995716e-06,
+      "loss": 0.9471,
+      "step": 230
+    },
+    {
+      "epoch": 0.49044585987261147,
+      "grad_norm": 0.6900933936326865,
+      "learning_rate": 5.613415719390759e-06,
+      "loss": 0.9058,
+      "step": 231
+    },
+    {
+      "epoch": 0.49256900212314225,
+      "grad_norm": 0.7383866221072647,
+      "learning_rate": 5.578525457121807e-06,
+      "loss": 1.0378,
+      "step": 232
+    },
+    {
+      "epoch": 0.49469214437367304,
+      "grad_norm": 0.732025682639867,
+      "learning_rate": 5.543606618593053e-06,
+      "loss": 0.9102,
+      "step": 233
+    },
+    {
+      "epoch": 0.4968152866242038,
+      "grad_norm": 0.7096362036491948,
+      "learning_rate": 5.508660928620216e-06,
+      "loss": 0.8952,
+      "step": 234
+    },
+    {
+      "epoch": 0.4989384288747346,
+      "grad_norm": 0.7099881315605873,
+      "learning_rate": 5.473690113345343e-06,
+      "loss": 0.9602,
+      "step": 235
+    },
+    {
+      "epoch": 0.5010615711252654,
+      "grad_norm": 0.6972671205744604,
+      "learning_rate": 5.438695900151537e-06,
+      "loss": 0.8624,
+      "step": 236
+    },
+    {
+      "epoch": 0.5031847133757962,
+      "grad_norm": 0.889396297433614,
+      "learning_rate": 5.403680017577653e-06,
+      "loss": 0.952,
+      "step": 237
+    },
+    {
+      "epoch": 0.505307855626327,
+      "grad_norm": 0.6402923695561287,
+      "learning_rate": 5.368644195232896e-06,
+      "loss": 0.9792,
+      "step": 238
+    },
+    {
+      "epoch": 0.5074309978768577,
+      "grad_norm": 0.6805565174215905,
+      "learning_rate": 5.3335901637113985e-06,
+      "loss": 0.9243,
+      "step": 239
+    },
+    {
+      "epoch": 0.5095541401273885,
+      "grad_norm": 0.6841794774774244,
+      "learning_rate": 5.298519654506736e-06,
+      "loss": 0.999,
+      "step": 240
+    },
+    {
+      "epoch": 0.5116772823779193,
+      "grad_norm": 0.7022154093220538,
+      "learning_rate": 5.2634343999263985e-06,
+      "loss": 0.9348,
+      "step": 241
+    },
+    {
+      "epoch": 0.5138004246284501,
+      "grad_norm": 0.6918338163749163,
+      "learning_rate": 5.228336133006223e-06,
+      "loss": 0.8975,
+      "step": 242
+    },
+    {
+      "epoch": 0.5159235668789809,
+      "grad_norm": 0.6755648031558511,
+      "learning_rate": 5.193226587424793e-06,
+      "loss": 0.9038,
+      "step": 243
+    },
+    {
+      "epoch": 0.5180467091295117,
+      "grad_norm": 0.7372704915893833,
+      "learning_rate": 5.158107497417795e-06,
+      "loss": 0.9826,
+      "step": 244
+    },
+    {
+      "epoch": 0.5201698513800425,
+      "grad_norm": 0.6463509797901226,
+      "learning_rate": 5.122980597692372e-06,
+      "loss": 0.9944,
+      "step": 245
+    },
+    {
+      "epoch": 0.5222929936305732,
+      "grad_norm": 0.6924850124684349,
+      "learning_rate": 5.087847623341421e-06,
+      "loss": 0.9763,
+      "step": 246
+    },
+    {
+      "epoch": 0.524416135881104,
+      "grad_norm": 0.6954437299483931,
+      "learning_rate": 5.052710309757899e-06,
+      "loss": 0.955,
+      "step": 247
+    },
+    {
+      "epoch": 0.5265392781316348,
+      "grad_norm": 0.6935825027209983,
+      "learning_rate": 5.0175703925490936e-06,
+      "loss": 0.9256,
+      "step": 248
+    },
+    {
+      "epoch": 0.5286624203821656,
+      "grad_norm": 0.691695101242488,
+      "learning_rate": 4.982429607450907e-06,
+      "loss": 0.9649,
+      "step": 249
+    },
+    {
+      "epoch": 0.5307855626326964,
+      "grad_norm": 0.6420506445010928,
+      "learning_rate": 4.947289690242103e-06,
+      "loss": 0.9767,
+      "step": 250
+    },
+    {
+      "epoch": 0.5329087048832272,
+      "grad_norm": 0.6983661270077597,
+      "learning_rate": 4.91215237665858e-06,
+      "loss": 0.9306,
+      "step": 251
+    },
+    {
+      "epoch": 0.535031847133758,
+      "grad_norm": 0.6859924966914502,
+      "learning_rate": 4.877019402307629e-06,
+      "loss": 0.9356,
+      "step": 252
+    },
+    {
+      "epoch": 0.5371549893842887,
+      "grad_norm": 0.7141495883745664,
+      "learning_rate": 4.841892502582206e-06,
+      "loss": 0.9602,
+      "step": 253
+    },
+    {
+      "epoch": 0.5392781316348195,
+      "grad_norm": 0.627839886124841,
+      "learning_rate": 4.806773412575211e-06,
+      "loss": 1.0012,
+      "step": 254
+    },
+    {
+      "epoch": 0.5414012738853503,
+      "grad_norm": 0.669446899340891,
+      "learning_rate": 4.7716638669937784e-06,
+      "loss": 0.9229,
+      "step": 255
+    },
+    {
+      "epoch": 0.5435244161358811,
+      "grad_norm": 0.681235443960255,
+      "learning_rate": 4.736565600073602e-06,
+      "loss": 0.9171,
+      "step": 256
+    },
+    {
+      "epoch": 0.5456475583864119,
+      "grad_norm": 0.7330756875715496,
+      "learning_rate": 4.701480345493266e-06,
+      "loss": 0.9257,
+      "step": 257
+    },
+    {
+      "epoch": 0.5477707006369427,
+      "grad_norm": 0.6630144658382536,
+      "learning_rate": 4.666409836288603e-06,
+      "loss": 0.9659,
+      "step": 258
+    },
+    {
+      "epoch": 0.5498938428874734,
+      "grad_norm": 0.7127193347963714,
+      "learning_rate": 4.631355804767106e-06,
+      "loss": 0.9375,
+      "step": 259
+    },
+    {
+      "epoch": 0.5520169851380042,
+      "grad_norm": 0.6290140804468636,
+      "learning_rate": 4.596319982422348e-06,
+      "loss": 0.9092,
+      "step": 260
+    },
+    {
+      "epoch": 0.554140127388535,
+      "grad_norm": 0.6986313542347127,
+      "learning_rate": 4.561304099848464e-06,
+      "loss": 0.9501,
+      "step": 261
+    },
+    {
+      "epoch": 0.5562632696390658,
+      "grad_norm": 0.6933713127536184,
+      "learning_rate": 4.526309886654659e-06,
+      "loss": 0.9977,
+      "step": 262
+    },
+    {
+      "epoch": 0.5583864118895966,
+      "grad_norm": 0.640132903868265,
+      "learning_rate": 4.491339071379783e-06,
+      "loss": 0.8758,
+      "step": 263
+    },
+    {
+      "epoch": 0.5605095541401274,
+      "grad_norm": 0.6822814887433616,
+      "learning_rate": 4.4563933814069475e-06,
+      "loss": 0.9315,
+      "step": 264
+    },
+    {
+      "epoch": 0.5626326963906582,
+      "grad_norm": 0.6398389559546622,
+      "learning_rate": 4.4214745428781946e-06,
+      "loss": 0.9847,
+      "step": 265
+    },
+    {
+      "epoch": 0.564755838641189,
+      "grad_norm": 0.6818041476558421,
+      "learning_rate": 4.386584280609242e-06,
+      "loss": 0.9819,
+      "step": 266
+    },
+    {
+      "epoch": 0.5668789808917197,
+      "grad_norm": 0.6679866325125441,
+      "learning_rate": 4.351724318004286e-06,
+      "loss": 0.9406,
+      "step": 267
+    },
+    {
+      "epoch": 0.5690021231422505,
+      "grad_norm": 0.6240664886135201,
+      "learning_rate": 4.316896376970866e-06,
+      "loss": 1.0318,
+      "step": 268
+    },
+    {
+      "epoch": 0.5711252653927813,
+      "grad_norm": 0.6220122266738475,
+      "learning_rate": 4.282102177834822e-06,
+      "loss": 0.9309,
+      "step": 269
+    },
+    {
+      "epoch": 0.5732484076433121,
+      "grad_norm": 0.6519539099798459,
+      "learning_rate": 4.2473434392553115e-06,
+      "loss": 0.9443,
+      "step": 270
+    },
+    {
+      "epoch": 0.5753715498938429,
+      "grad_norm": 0.6194962484361002,
+      "learning_rate": 4.212621878139912e-06,
+      "loss": 0.9332,
+      "step": 271
+    },
+    {
+      "epoch": 0.5774946921443737,
+      "grad_norm": 0.6484200470987919,
+      "learning_rate": 4.177939209559828e-06,
+      "loss": 0.9875,
+      "step": 272
+    },
+    {
+      "epoch": 0.5796178343949044,
+      "grad_norm": 0.6330847653688858,
+      "learning_rate": 4.143297146665167e-06,
+      "loss": 0.9074,
+      "step": 273
+    },
+    {
+      "epoch": 0.5817409766454352,
+      "grad_norm": 0.6413164774139387,
+      "learning_rate": 4.108697400600316e-06,
+      "loss": 0.9854,
+      "step": 274
+    },
+    {
+      "epoch": 0.583864118895966,
+      "grad_norm": 0.6779890723404448,
+      "learning_rate": 4.074141680419422e-06,
+      "loss": 0.9924,
+      "step": 275
+    },
+    {
+      "epoch": 0.5859872611464968,
+      "grad_norm": 0.7145362735963939,
+      "learning_rate": 4.039631693001976e-06,
+      "loss": 0.9673,
+      "step": 276
+    },
+    {
+      "epoch": 0.5881104033970276,
+      "grad_norm": 0.6462201481263671,
+      "learning_rate": 4.005169142968503e-06,
+      "loss": 0.9365,
+      "step": 277
+    },
+    {
+      "epoch": 0.5902335456475584,
+      "grad_norm": 0.6400041384015616,
+      "learning_rate": 3.970755732596349e-06,
+      "loss": 0.9345,
+      "step": 278
+    },
+    {
+      "epoch": 0.5923566878980892,
+      "grad_norm": 0.6705732923411916,
+      "learning_rate": 3.936393161735616e-06,
+      "loss": 0.9745,
+      "step": 279
+    },
+    {
+      "epoch": 0.5944798301486199,
+      "grad_norm": 0.6076428308867137,
+      "learning_rate": 3.902083127725186e-06,
+      "loss": 0.9887,
+      "step": 280
+    },
+    {
+      "epoch": 0.5966029723991507,
+      "grad_norm": 0.6829854410878823,
+      "learning_rate": 3.867827325308882e-06,
+      "loss": 0.9788,
+      "step": 281
+    },
+    {
+      "epoch": 0.5987261146496815,
+      "grad_norm": 0.635794721170443,
+      "learning_rate": 3.83362744655176e-06,
+      "loss": 0.9084,
+      "step": 282
+    },
+    {
+      "epoch": 0.6008492569002123,
+      "grad_norm": 0.6406247555663112,
+      "learning_rate": 3.799485180756526e-06,
+      "loss": 0.9696,
+      "step": 283
+    },
+    {
+      "epoch": 0.6029723991507431,
+      "grad_norm": 0.6836159898583682,
+      "learning_rate": 3.765402214380095e-06,
+      "loss": 0.9622,
+      "step": 284
+    },
+    {
+      "epoch": 0.6050955414012739,
+      "grad_norm": 0.6903265273050954,
+      "learning_rate": 3.731380230950288e-06,
+      "loss": 0.9804,
+      "step": 285
+    },
+    {
+      "epoch": 0.6072186836518046,
+      "grad_norm": 0.6582988195251387,
+      "learning_rate": 3.6974209109826724e-06,
+      "loss": 0.9198,
+      "step": 286
+    },
+    {
+      "epoch": 0.6093418259023354,
+      "grad_norm": 0.616448368616492,
+      "learning_rate": 3.663525931897559e-06,
+      "loss": 0.9158,
+      "step": 287
+    },
+    {
+      "epoch": 0.6114649681528662,
+      "grad_norm": 0.6673765701597619,
+      "learning_rate": 3.6296969679371325e-06,
+      "loss": 0.9611,
+      "step": 288
+    },
+    {
+      "epoch": 0.613588110403397,
+      "grad_norm": 0.6888915388418313,
+      "learning_rate": 3.595935690082769e-06,
+      "loss": 0.8919,
+      "step": 289
+    },
+    {
+      "epoch": 0.6157112526539278,
+      "grad_norm": 0.6101159334044771,
+      "learning_rate": 3.56224376597249e-06,
+      "loss": 0.9515,
+      "step": 290
+    },
+    {
+      "epoch": 0.6178343949044586,
+      "grad_norm": 0.658669764640868,
+      "learning_rate": 3.528622859818582e-06,
+      "loss": 0.9566,
+      "step": 291
+    },
+    {
+      "epoch": 0.6199575371549894,
+      "grad_norm": 0.5954645751379325,
+      "learning_rate": 3.495074632325407e-06,
+      "loss": 0.9686,
+      "step": 292
+    },
+    {
+      "epoch": 0.6220806794055201,
+      "grad_norm": 0.6417440985880237,
+      "learning_rate": 3.461600740607366e-06,
+      "loss": 0.9841,
+      "step": 293
+    },
+    {
+      "epoch": 0.6242038216560509,
+      "grad_norm": 0.6513493598293557,
+      "learning_rate": 3.4282028381070366e-06,
+      "loss": 0.88,
+      "step": 294
+    },
+    {
+      "epoch": 0.6263269639065817,
+      "grad_norm": 0.6460724764075341,
+      "learning_rate": 3.3948825745135196e-06,
+      "loss": 0.9395,
+      "step": 295
+    },
+    {
+      "epoch": 0.6284501061571125,
+      "grad_norm": 0.7273038274491741,
+      "learning_rate": 3.361641595680937e-06,
+      "loss": 0.9796,
+      "step": 296
+    },
+    {
+      "epoch": 0.6305732484076433,
+      "grad_norm": 0.63867402479588,
+      "learning_rate": 3.3284815435471423e-06,
+      "loss": 0.9616,
+      "step": 297
+    },
+    {
+      "epoch": 0.6326963906581741,
+      "grad_norm": 0.6249901633020171,
+      "learning_rate": 3.295404056052616e-06,
+      "loss": 0.9268,
+      "step": 298
+    },
+    {
+      "epoch": 0.6348195329087049,
+      "grad_norm": 0.7130471341801802,
+      "learning_rate": 3.2624107670595567e-06,
+      "loss": 0.97,
+      "step": 299
+    },
+    {
+      "epoch": 0.6369426751592356,
+      "grad_norm": 0.6679078460787682,
+      "learning_rate": 3.2295033062711823e-06,
+      "loss": 0.9088,
+      "step": 300
+    },
+    {
+      "epoch": 0.6390658174097664,
+      "grad_norm": 0.7012208037782451,
+      "learning_rate": 3.1966832991512232e-06,
+      "loss": 0.9859,
+      "step": 301
+    },
+    {
+      "epoch": 0.6411889596602972,
+      "grad_norm": 0.6533943769965103,
+      "learning_rate": 3.16395236684364e-06,
+      "loss": 0.9511,
+      "step": 302
+    },
+    {
+      "epoch": 0.643312101910828,
+      "grad_norm": 0.682194483085267,
+      "learning_rate": 3.131312126092544e-06,
+      "loss": 0.9403,
+      "step": 303
+    },
+    {
+      "epoch": 0.6454352441613588,
+      "grad_norm": 0.6703363053760469,
+      "learning_rate": 3.098764189162332e-06,
+      "loss": 0.9039,
+      "step": 304
+    },
+    {
+      "epoch": 0.6475583864118896,
+      "grad_norm": 0.6294417096433408,
+      "learning_rate": 3.0663101637580626e-06,
+      "loss": 0.9244,
+      "step": 305
+    },
+    {
+      "epoch": 0.6496815286624203,
+      "grad_norm": 0.6602293200539316,
+      "learning_rate": 3.03395165294603e-06,
+      "loss": 0.9533,
+      "step": 306
+    },
+    {
+      "epoch": 0.6518046709129511,
+      "grad_norm": 0.672086498728235,
+      "learning_rate": 3.0016902550745896e-06,
+      "loss": 0.953,
+      "step": 307
+    },
+    {
+      "epoch": 0.6539278131634819,
+      "grad_norm": 0.6146168409218723,
+      "learning_rate": 2.9695275636951983e-06,
+      "loss": 0.9569,
+      "step": 308
+    },
+    {
+      "epoch": 0.6560509554140127,
+      "grad_norm": 0.6219654134664444,
+      "learning_rate": 2.9374651674837128e-06,
+      "loss": 1.0229,
+      "step": 309
+    },
+    {
+      "epoch": 0.6581740976645435,
+      "grad_norm": 0.6353930075528244,
+      "learning_rate": 2.9055046501619088e-06,
+      "loss": 0.9792,
+      "step": 310
+    },
+    {
+      "epoch": 0.6602972399150743,
+      "grad_norm": 0.7205625550852138,
+      "learning_rate": 2.8736475904192516e-06,
+      "loss": 0.9297,
+      "step": 311
+    },
+    {
+      "epoch": 0.6624203821656051,
+      "grad_norm": 0.65227066914161,
+      "learning_rate": 2.841895561834927e-06,
+      "loss": 0.8995,
+      "step": 312
+    },
+    {
+      "epoch": 0.6645435244161358,
+      "grad_norm": 0.6736462207854454,
+      "learning_rate": 2.810250132800103e-06,
+      "loss": 0.9456,
+      "step": 313
+    },
+    {
+      "epoch": 0.6666666666666666,
+      "grad_norm": 0.6024110164676963,
+      "learning_rate": 2.778712866440464e-06,
+      "loss": 0.9641,
+      "step": 314
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 471,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 157,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 46655615139840.0,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b034e0d83300ea1c9e38edcef801137d8d09a7d4a653638cef26385c245e0d4
+size 8017

zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/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 . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+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, weights_only=False)
+
+        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 tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # 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}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] 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")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+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 = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * 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
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        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}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        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 to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=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
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    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.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    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")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               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_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``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
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+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_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    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_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)