Upload model checkpoint v1

.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

chat_template.jinja

@@ -0,0 +1,47 @@
+{{ bos_token }}
+{%- if messages[0]['role'] == 'system' -%}
+    {%- if messages[0]['content'] is string -%}
+        {%- set first_user_prefix = messages[0]['content'] + '
+
+' -%}
+    {%- else -%}
+        {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
+
+' -%}
+    {%- endif -%}
+    {%- set loop_messages = messages[1:] -%}
+{%- else -%}
+    {%- set first_user_prefix = "" -%}
+    {%- set loop_messages = messages -%}
+{%- endif -%}
+{%- for message in loop_messages -%}
+    {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
+        {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
+    {%- endif -%}
+    {%- if (message['role'] == 'assistant') -%}
+        {%- set role = "model" -%}
+    {%- else -%}
+        {%- set role = message['role'] -%}
+    {%- endif -%}
+    {{ '<start_of_turn>' + role + '
+' + (first_user_prefix if loop.first else "") }}
+    {%- if message['content'] is string -%}
+        {{ message['content'] | trim }}
+    {%- elif message['content'] is iterable -%}
+        {%- for item in message['content'] -%}
+            {%- if item['type'] == 'image' -%}
+                {{ '<start_of_image>' }}
+            {%- elif item['type'] == 'text' -%}
+                {{ item['text'] | trim }}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{ raise_exception("Invalid content type") }}
+    {%- endif -%}
+    {{ '<end_of_turn>
+' }}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{'<start_of_turn>model
+'}}
+{%- endif -%}

config.json

@@ -0,0 +1,93 @@
+{
+  "_sliding_window_pattern": 6,
+  "architectures": [
+    "Gemma3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attn_logit_softcapping": null,
+  "bos_token_id": 2,
+  "dtype": "bfloat16",
+  "eos_token_id": 106,
+  "final_logit_softcapping": null,
+  "head_dim": 256,
+  "hidden_activation": "gelu_pytorch_tanh",
+  "hidden_size": 3840,
+  "initializer_range": 0.02,
+  "intermediate_size": 15360,
+  "layer_types": [
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 131072,
+  "model_type": "gemma3_text",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 48,
+  "num_key_value_heads": 8,
+  "pad_token_id": 0,
+  "query_pre_attn_scalar": 256,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "full_attention": {
+      "factor": 8.0,
+      "rope_theta": 1000000.0,
+      "rope_type": "linear"
+    },
+    "sliding_attention": {
+      "rope_theta": 10000.0,
+      "rope_type": "default"
+    }
+  },
+  "sliding_window": 1024,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.6.0",
+  "use_bidirectional_attention": false,
+  "use_cache": false,
+  "vocab_size": 262208
+}

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 2,
+  "eos_token_id": [
+    106,
+    1
+  ],
+  "pad_token_id": 0,
+  "transformers_version": "5.6.0"
+}

latest

@@ -0,0 +1 @@
+global_step1000

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c882cadcc07b90e6faf5b85b2478a412739d8f61191ced30a263d7dfce8b6e17
+size 23532142208

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:022e5f570f01a12c1d81b42bd6eb19a0e173e7f47e424219366d42929f99e132
+size 15429

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3aed2436bc3ad431adb01e8d4df32815773aab4a6b13c3adb4fced6192268c3
+size 15429

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97dd34720bb0b20ed7301d9f60cd440d867fb1aac58f9dc15c1f7fe4ac70e520
+size 15429

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37832d386fc739839d36940cdc0c137876018d931548a6ee584a9d52680e1a59
+size 15429

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bf78d3f02d3c1bc757679748d981cc5a592427f5c20b94459c0be2999b97072
+size 1465

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daab2354f8a74e70d70b4d1f804939b68a8c9624dd06cb7858e52dd8970e9726
+size 33384567

tokenizer_config.json

@@ -0,0 +1,29 @@
+{
+  "backend": "tokenizers",
+  "boi_token": "<start_of_image>",
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "<bos>",
+  "eoi_token": "<end_of_image>",
+  "eos_token": "<end_of_turn>",
+  "image_token": "<image_soft_token>",
+  "is_local": false,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "model_specific_special_tokens": {
+    "boi_token": "<start_of_image>",
+    "eoi_token": "<end_of_image>",
+    "image_token": "<image_soft_token>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "processor_class": "Gemma3Processor",
+  "sep_token": "<eos>",
+  "sp_model_kwargs": null,
+  "spaces_between_special_tokens": false,
+  "split_special_tokens": false,
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

trainer_state.json

@@ -0,0 +1,1434 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.700985761226725,
+  "eval_steps": 500,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.04381161007667032,
+      "grad_norm": 15.216287445196276,
+      "learning_rate": 8e-09,
+      "loss": 0.48418779373168946,
+      "step": 5
+    },
+    {
+      "epoch": 0.08762322015334063,
+      "grad_norm": 23.36358717964335,
+      "learning_rate": 1.8e-08,
+      "loss": 0.46855783462524414,
+      "step": 10
+    },
+    {
+      "epoch": 0.13143483023001096,
+      "grad_norm": 20.000826258755275,
+      "learning_rate": 2.8e-08,
+      "loss": 0.47519798278808595,
+      "step": 15
+    },
+    {
+      "epoch": 0.17524644030668127,
+      "grad_norm": 15.69050377411594,
+      "learning_rate": 3.7999999999999996e-08,
+      "loss": 0.4735735893249512,
+      "step": 20
+    },
+    {
+      "epoch": 0.21905805038335158,
+      "grad_norm": 16.238081490850117,
+      "learning_rate": 4.8e-08,
+      "loss": 0.4729929447174072,
+      "step": 25
+    },
+    {
+      "epoch": 0.2628696604600219,
+      "grad_norm": 17.49951407045111,
+      "learning_rate": 5.8e-08,
+      "loss": 0.47327580451965334,
+      "step": 30
+    },
+    {
+      "epoch": 0.3066812705366922,
+      "grad_norm": 26.57129804217329,
+      "learning_rate": 6.8e-08,
+      "loss": 0.47186527252197263,
+      "step": 35
+    },
+    {
+      "epoch": 0.35049288061336253,
+      "grad_norm": 20.11949989372027,
+      "learning_rate": 7.8e-08,
+      "loss": 0.4615940093994141,
+      "step": 40
+    },
+    {
+      "epoch": 0.39430449069003287,
+      "grad_norm": 64.31215127356752,
+      "learning_rate": 8.8e-08,
+      "loss": 0.4518611431121826,
+      "step": 45
+    },
+    {
+      "epoch": 0.43811610076670315,
+      "grad_norm": 15.857571041423546,
+      "learning_rate": 9.8e-08,
+      "loss": 0.43337259292602537,
+      "step": 50
+    },
+    {
+      "epoch": 0.4819277108433735,
+      "grad_norm": 12.436173813521455,
+      "learning_rate": 1.0799999999999999e-07,
+      "loss": 0.41027131080627444,
+      "step": 55
+    },
+    {
+      "epoch": 0.5257393209200438,
+      "grad_norm": 12.565891289933171,
+      "learning_rate": 1.1799999999999998e-07,
+      "loss": 0.40830144882202146,
+      "step": 60
+    },
+    {
+      "epoch": 0.5695509309967142,
+      "grad_norm": 8.403712696803929,
+      "learning_rate": 1.28e-07,
+      "loss": 0.3690589427947998,
+      "step": 65
+    },
+    {
+      "epoch": 0.6133625410733844,
+      "grad_norm": 5.488688842051114,
+      "learning_rate": 1.3800000000000002e-07,
+      "loss": 0.33161895275115966,
+      "step": 70
+    },
+    {
+      "epoch": 0.6571741511500547,
+      "grad_norm": 4.750780287029476,
+      "learning_rate": 1.4799999999999998e-07,
+      "loss": 0.31440370082855223,
+      "step": 75
+    },
+    {
+      "epoch": 0.7009857612267251,
+      "grad_norm": 5.694357008951302,
+      "learning_rate": 1.5799999999999999e-07,
+      "loss": 0.29302048683166504,
+      "step": 80
+    },
+    {
+      "epoch": 0.7447973713033954,
+      "grad_norm": 3.6650496363290634,
+      "learning_rate": 1.68e-07,
+      "loss": 0.26142473220825196,
+      "step": 85
+    },
+    {
+      "epoch": 0.7886089813800657,
+      "grad_norm": 2.8616266624672777,
+      "learning_rate": 1.7799999999999998e-07,
+      "loss": 0.2410365104675293,
+      "step": 90
+    },
+    {
+      "epoch": 0.8324205914567361,
+      "grad_norm": 2.0793907207095534,
+      "learning_rate": 1.88e-07,
+      "loss": 0.2225107431411743,
+      "step": 95
+    },
+    {
+      "epoch": 0.8762322015334063,
+      "grad_norm": 2.771283288158846,
+      "learning_rate": 1.98e-07,
+      "loss": 0.21108705997467042,
+      "step": 100
+    },
+    {
+      "epoch": 0.9200438116100766,
+      "grad_norm": 1.913027569576905,
+      "learning_rate": 2.0799999999999998e-07,
+      "loss": 0.20774705410003663,
+      "step": 105
+    },
+    {
+      "epoch": 0.963855421686747,
+      "grad_norm": 1.4521809664515546,
+      "learning_rate": 2.18e-07,
+      "loss": 0.1867969274520874,
+      "step": 110
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 2.4658234086980677,
+      "learning_rate": 2.28e-07,
+      "loss": 0.1816505789756775,
+      "step": 115
+    },
+    {
+      "epoch": 1.0438116100766703,
+      "grad_norm": 1.5311397736316652,
+      "learning_rate": 2.38e-07,
+      "loss": 0.16964107751846313,
+      "step": 120
+    },
+    {
+      "epoch": 1.0876232201533407,
+      "grad_norm": 3.290774967566002,
+      "learning_rate": 2.48e-07,
+      "loss": 0.15306370258331298,
+      "step": 125
+    },
+    {
+      "epoch": 1.131434830230011,
+      "grad_norm": 0.8940101199085605,
+      "learning_rate": 2.58e-07,
+      "loss": 0.14721683263778687,
+      "step": 130
+    },
+    {
+      "epoch": 1.1752464403066814,
+      "grad_norm": 1.5394515578260013,
+      "learning_rate": 2.68e-07,
+      "loss": 0.15360240936279296,
+      "step": 135
+    },
+    {
+      "epoch": 1.2190580503833517,
+      "grad_norm": 1.7858847630309493,
+      "learning_rate": 2.7800000000000003e-07,
+      "loss": 0.15086040496826172,
+      "step": 140
+    },
+    {
+      "epoch": 1.262869660460022,
+      "grad_norm": 0.8019861150580353,
+      "learning_rate": 2.88e-07,
+      "loss": 0.14722588062286376,
+      "step": 145
+    },
+    {
+      "epoch": 1.3066812705366921,
+      "grad_norm": 1.0183119548120452,
+      "learning_rate": 2.98e-07,
+      "loss": 0.1381218194961548,
+      "step": 150
+    },
+    {
+      "epoch": 1.3504928806133625,
+      "grad_norm": 0.7483621942757548,
+      "learning_rate": 3.08e-07,
+      "loss": 0.14110673666000367,
+      "step": 155
+    },
+    {
+      "epoch": 1.3943044906900328,
+      "grad_norm": 0.7139120208182828,
+      "learning_rate": 3.18e-07,
+      "loss": 0.14381901025772095,
+      "step": 160
+    },
+    {
+      "epoch": 1.4381161007667032,
+      "grad_norm": 0.5818206370662435,
+      "learning_rate": 3.28e-07,
+      "loss": 0.13782813549041747,
+      "step": 165
+    },
+    {
+      "epoch": 1.4819277108433735,
+      "grad_norm": 0.5922737231474884,
+      "learning_rate": 3.38e-07,
+      "loss": 0.14187902212142944,
+      "step": 170
+    },
+    {
+      "epoch": 1.5257393209200438,
+      "grad_norm": 0.5907207617335364,
+      "learning_rate": 3.4799999999999994e-07,
+      "loss": 0.13219969272613524,
+      "step": 175
+    },
+    {
+      "epoch": 1.5695509309967142,
+      "grad_norm": 0.7395669120270896,
+      "learning_rate": 3.5799999999999995e-07,
+      "loss": 0.13412785530090332,
+      "step": 180
+    },
+    {
+      "epoch": 1.6133625410733843,
+      "grad_norm": 0.7389843099563854,
+      "learning_rate": 3.6799999999999996e-07,
+      "loss": 0.13302382230758666,
+      "step": 185
+    },
+    {
+      "epoch": 1.6571741511500546,
+      "grad_norm": 0.46928725746320077,
+      "learning_rate": 3.7799999999999997e-07,
+      "loss": 0.12532199621200563,
+      "step": 190
+    },
+    {
+      "epoch": 1.700985761226725,
+      "grad_norm": 1.4378334177577896,
+      "learning_rate": 3.88e-07,
+      "loss": 0.12993998527526857,
+      "step": 195
+    },
+    {
+      "epoch": 1.7447973713033953,
+      "grad_norm": 1.18829939619434,
+      "learning_rate": 3.98e-07,
+      "loss": 0.1237565279006958,
+      "step": 200
+    },
+    {
+      "epoch": 1.7886089813800656,
+      "grad_norm": 0.5041162661955574,
+      "learning_rate": 4.0799999999999995e-07,
+      "loss": 0.1410191297531128,
+      "step": 205
+    },
+    {
+      "epoch": 1.832420591456736,
+      "grad_norm": 0.5263127851191748,
+      "learning_rate": 4.1799999999999996e-07,
+      "loss": 0.12495659589767456,
+      "step": 210
+    },
+    {
+      "epoch": 1.8762322015334063,
+      "grad_norm": 0.45393764404596665,
+      "learning_rate": 4.2799999999999997e-07,
+      "loss": 0.12057209014892578,
+      "step": 215
+    },
+    {
+      "epoch": 1.9200438116100766,
+      "grad_norm": 0.5034037682395058,
+      "learning_rate": 4.38e-07,
+      "loss": 0.12823106050491334,
+      "step": 220
+    },
+    {
+      "epoch": 1.963855421686747,
+      "grad_norm": 0.4188960855911732,
+      "learning_rate": 4.48e-07,
+      "loss": 0.11971760988235473,
+      "step": 225
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.892183746093029,
+      "learning_rate": 4.58e-07,
+      "loss": 0.11903363466262817,
+      "step": 230
+    },
+    {
+      "epoch": 2.0438116100766703,
+      "grad_norm": 0.8417597566788897,
+      "learning_rate": 4.68e-07,
+      "loss": 0.1283984065055847,
+      "step": 235
+    },
+    {
+      "epoch": 2.0876232201533407,
+      "grad_norm": 0.6569068480529452,
+      "learning_rate": 4.779999999999999e-07,
+      "loss": 0.12451218366622925,
+      "step": 240
+    },
+    {
+      "epoch": 2.131434830230011,
+      "grad_norm": 1.9369437227625623,
+      "learning_rate": 4.879999999999999e-07,
+      "loss": 0.12201762199401855,
+      "step": 245
+    },
+    {
+      "epoch": 2.1752464403066814,
+      "grad_norm": 0.492143785556789,
+      "learning_rate": 4.979999999999999e-07,
+      "loss": 0.11702324151992798,
+      "step": 250
+    },
+    {
+      "epoch": 2.2190580503833517,
+      "grad_norm": 0.9985585601716669,
+      "learning_rate": 5.079999999999999e-07,
+      "loss": 0.11661801338195801,
+      "step": 255
+    },
+    {
+      "epoch": 2.262869660460022,
+      "grad_norm": 0.4986662652452154,
+      "learning_rate": 5.18e-07,
+      "loss": 0.12300481796264648,
+      "step": 260
+    },
+    {
+      "epoch": 2.3066812705366924,
+      "grad_norm": 0.42171845691403576,
+      "learning_rate": 5.28e-07,
+      "loss": 0.12354564666748047,
+      "step": 265
+    },
+    {
+      "epoch": 2.3504928806133627,
+      "grad_norm": 1.0407715899751926,
+      "learning_rate": 5.38e-07,
+      "loss": 0.12628095149993895,
+      "step": 270
+    },
+    {
+      "epoch": 2.394304490690033,
+      "grad_norm": 0.3591569287617796,
+      "learning_rate": 5.48e-07,
+      "loss": 0.12059959173202514,
+      "step": 275
+    },
+    {
+      "epoch": 2.4381161007667034,
+      "grad_norm": 0.34709206358048095,
+      "learning_rate": 5.58e-07,
+      "loss": 0.12300305366516114,
+      "step": 280
+    },
+    {
+      "epoch": 2.4819277108433733,
+      "grad_norm": 0.3267656852125577,
+      "learning_rate": 5.679999999999999e-07,
+      "loss": 0.11873151063919067,
+      "step": 285
+    },
+    {
+      "epoch": 2.525739320920044,
+      "grad_norm": 0.36315891368762354,
+      "learning_rate": 5.779999999999999e-07,
+      "loss": 0.11392134428024292,
+      "step": 290
+    },
+    {
+      "epoch": 2.569550930996714,
+      "grad_norm": 0.401643325163856,
+      "learning_rate": 5.879999999999999e-07,
+      "loss": 0.11619458198547364,
+      "step": 295
+    },
+    {
+      "epoch": 2.6133625410733843,
+      "grad_norm": 0.5780107770877003,
+      "learning_rate": 5.979999999999999e-07,
+      "loss": 0.11880600452423096,
+      "step": 300
+    },
+    {
+      "epoch": 2.6571741511500546,
+      "grad_norm": 0.3511477760975858,
+      "learning_rate": 6.079999999999999e-07,
+      "loss": 0.11613132953643798,
+      "step": 305
+    },
+    {
+      "epoch": 2.700985761226725,
+      "grad_norm": 0.4629256939960936,
+      "learning_rate": 6.18e-07,
+      "loss": 0.11528429985046387,
+      "step": 310
+    },
+    {
+      "epoch": 2.7447973713033953,
+      "grad_norm": 0.46045542772623654,
+      "learning_rate": 6.28e-07,
+      "loss": 0.11843544244766235,
+      "step": 315
+    },
+    {
+      "epoch": 2.7886089813800656,
+      "grad_norm": 0.37854915120496596,
+      "learning_rate": 6.38e-07,
+      "loss": 0.11221380233764648,
+      "step": 320
+    },
+    {
+      "epoch": 2.832420591456736,
+      "grad_norm": 0.44957650893043816,
+      "learning_rate": 6.48e-07,
+      "loss": 0.12133526802062988,
+      "step": 325
+    },
+    {
+      "epoch": 2.8762322015334063,
+      "grad_norm": 0.3963359279128714,
+      "learning_rate": 6.58e-07,
+      "loss": 0.11056663990020751,
+      "step": 330
+    },
+    {
+      "epoch": 2.9200438116100766,
+      "grad_norm": 0.33624224172102196,
+      "learning_rate": 6.68e-07,
+      "loss": 0.1102494478225708,
+      "step": 335
+    },
+    {
+      "epoch": 2.963855421686747,
+      "grad_norm": 0.4982044847656389,
+      "learning_rate": 6.78e-07,
+      "loss": 0.11427903175354004,
+      "step": 340
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 1.2089952315519579,
+      "learning_rate": 6.879999999999999e-07,
+      "loss": 0.11158945560455322,
+      "step": 345
+    },
+    {
+      "epoch": 3.0438116100766703,
+      "grad_norm": 0.38194453310959764,
+      "learning_rate": 6.979999999999999e-07,
+      "loss": 0.11192548274993896,
+      "step": 350
+    },
+    {
+      "epoch": 3.0876232201533407,
+      "grad_norm": 0.440842233975557,
+      "learning_rate": 7.079999999999999e-07,
+      "loss": 0.11824355125427247,
+      "step": 355
+    },
+    {
+      "epoch": 3.131434830230011,
+      "grad_norm": 0.39451329579453837,
+      "learning_rate": 7.179999999999999e-07,
+      "loss": 0.11076958179473877,
+      "step": 360
+    },
+    {
+      "epoch": 3.1752464403066814,
+      "grad_norm": 0.46024551490235244,
+      "learning_rate": 7.28e-07,
+      "loss": 0.11353641748428345,
+      "step": 365
+    },
+    {
+      "epoch": 3.2190580503833517,
+      "grad_norm": 0.41036470861070845,
+      "learning_rate": 7.38e-07,
+      "loss": 0.1143984079360962,
+      "step": 370
+    },
+    {
+      "epoch": 3.262869660460022,
+      "grad_norm": 0.3775717804357684,
+      "learning_rate": 7.48e-07,
+      "loss": 0.10811513662338257,
+      "step": 375
+    },
+    {
+      "epoch": 3.3066812705366924,
+      "grad_norm": 0.5708506609199996,
+      "learning_rate": 7.58e-07,
+      "loss": 0.11109951734542847,
+      "step": 380
+    },
+    {
+      "epoch": 3.3504928806133627,
+      "grad_norm": 1.1959381812639907,
+      "learning_rate": 7.68e-07,
+      "loss": 0.11139968633651734,
+      "step": 385
+    },
+    {
+      "epoch": 3.394304490690033,
+      "grad_norm": 0.4199932705678528,
+      "learning_rate": 7.78e-07,
+      "loss": 0.11237516403198242,
+      "step": 390
+    },
+    {
+      "epoch": 3.4381161007667034,
+      "grad_norm": 0.8905765883619894,
+      "learning_rate": 7.88e-07,
+      "loss": 0.1078277587890625,
+      "step": 395
+    },
+    {
+      "epoch": 3.4819277108433733,
+      "grad_norm": 0.597433622094112,
+      "learning_rate": 7.98e-07,
+      "loss": 0.11192183494567871,
+      "step": 400
+    },
+    {
+      "epoch": 3.525739320920044,
+      "grad_norm": 0.3597039632212566,
+      "learning_rate": 8.08e-07,
+      "loss": 0.10791292190551757,
+      "step": 405
+    },
+    {
+      "epoch": 3.569550930996714,
+      "grad_norm": 0.44603735834013325,
+      "learning_rate": 8.179999999999999e-07,
+      "loss": 0.11222789287567139,
+      "step": 410
+    },
+    {
+      "epoch": 3.6133625410733843,
+      "grad_norm": 0.4145582397207868,
+      "learning_rate": 8.28e-07,
+      "loss": 0.1124007225036621,
+      "step": 415
+    },
+    {
+      "epoch": 3.6571741511500546,
+      "grad_norm": 0.4475638617058225,
+      "learning_rate": 8.38e-07,
+      "loss": 0.11902028322219849,
+      "step": 420
+    },
+    {
+      "epoch": 3.700985761226725,
+      "grad_norm": 0.41336046966692364,
+      "learning_rate": 8.48e-07,
+      "loss": 0.10512697696685791,
+      "step": 425
+    },
+    {
+      "epoch": 3.7447973713033953,
+      "grad_norm": 0.395369572152183,
+      "learning_rate": 8.58e-07,
+      "loss": 0.1127134084701538,
+      "step": 430
+    },
+    {
+      "epoch": 3.7886089813800656,
+      "grad_norm": 0.37821864648028486,
+      "learning_rate": 8.68e-07,
+      "loss": 0.10732884407043457,
+      "step": 435
+    },
+    {
+      "epoch": 3.832420591456736,
+      "grad_norm": 0.4320274336328645,
+      "learning_rate": 8.78e-07,
+      "loss": 0.11337308883666992,
+      "step": 440
+    },
+    {
+      "epoch": 3.8762322015334063,
+      "grad_norm": 0.5384978814732388,
+      "learning_rate": 8.88e-07,
+      "loss": 0.10675361156463622,
+      "step": 445
+    },
+    {
+      "epoch": 3.9200438116100766,
+      "grad_norm": 0.4986740950497558,
+      "learning_rate": 8.98e-07,
+      "loss": 0.10964843034744262,
+      "step": 450
+    },
+    {
+      "epoch": 3.963855421686747,
+      "grad_norm": 0.6756114047123225,
+      "learning_rate": 9.08e-07,
+      "loss": 0.1119537353515625,
+      "step": 455
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.8222551505954045,
+      "learning_rate": 9.18e-07,
+      "loss": 0.11235996484756469,
+      "step": 460
+    },
+    {
+      "epoch": 4.04381161007667,
+      "grad_norm": 0.4521119513476695,
+      "learning_rate": 9.28e-07,
+      "loss": 0.1151043176651001,
+      "step": 465
+    },
+    {
+      "epoch": 4.087623220153341,
+      "grad_norm": 0.5086183420648134,
+      "learning_rate": 9.379999999999998e-07,
+      "loss": 0.10557041168212891,
+      "step": 470
+    },
+    {
+      "epoch": 4.131434830230011,
+      "grad_norm": 0.436912186073042,
+      "learning_rate": 9.479999999999999e-07,
+      "loss": 0.10844831466674805,
+      "step": 475
+    },
+    {
+      "epoch": 4.175246440306681,
+      "grad_norm": 0.386664941548771,
+      "learning_rate": 9.58e-07,
+      "loss": 0.10682735443115235,
+      "step": 480
+    },
+    {
+      "epoch": 4.219058050383351,
+      "grad_norm": 0.4069071493133063,
+      "learning_rate": 9.679999999999999e-07,
+      "loss": 0.11006146669387817,
+      "step": 485
+    },
+    {
+      "epoch": 4.262869660460022,
+      "grad_norm": 0.2861155321012741,
+      "learning_rate": 9.78e-07,
+      "loss": 0.10587197542190552,
+      "step": 490
+    },
+    {
+      "epoch": 4.306681270536692,
+      "grad_norm": 0.3742289134536855,
+      "learning_rate": 9.88e-07,
+      "loss": 0.11084235906600952,
+      "step": 495
+    },
+    {
+      "epoch": 4.350492880613363,
+      "grad_norm": 0.5664218032029411,
+      "learning_rate": 9.98e-07,
+      "loss": 0.10497891902923584,
+      "step": 500
+    },
+    {
+      "epoch": 4.394304490690033,
+      "grad_norm": 0.38678093362225613,
+      "learning_rate": 1e-06,
+      "loss": 0.10288872718811035,
+      "step": 505
+    },
+    {
+      "epoch": 4.438116100766703,
+      "grad_norm": 0.37721106516680397,
+      "learning_rate": 1e-06,
+      "loss": 0.10397158861160279,
+      "step": 510
+    },
+    {
+      "epoch": 4.481927710843373,
+      "grad_norm": 0.4241488499774428,
+      "learning_rate": 1e-06,
+      "loss": 0.10589138269424439,
+      "step": 515
+    },
+    {
+      "epoch": 4.525739320920044,
+      "grad_norm": 0.911082168966086,
+      "learning_rate": 1e-06,
+      "loss": 0.10726232528686523,
+      "step": 520
+    },
+    {
+      "epoch": 4.569550930996714,
+      "grad_norm": 0.6340378208983399,
+      "learning_rate": 1e-06,
+      "loss": 0.10499470233917237,
+      "step": 525
+    },
+    {
+      "epoch": 4.613362541073385,
+      "grad_norm": 0.4921809624463462,
+      "learning_rate": 1e-06,
+      "loss": 0.10593559741973876,
+      "step": 530
+    },
+    {
+      "epoch": 4.657174151150055,
+      "grad_norm": 0.6445598019057156,
+      "learning_rate": 1e-06,
+      "loss": 0.10545206069946289,
+      "step": 535
+    },
+    {
+      "epoch": 4.700985761226725,
+      "grad_norm": 0.6428232947903155,
+      "learning_rate": 1e-06,
+      "loss": 0.10749400854110717,
+      "step": 540
+    },
+    {
+      "epoch": 4.744797371303395,
+      "grad_norm": 0.3865700878305047,
+      "learning_rate": 1e-06,
+      "loss": 0.10601369142532349,
+      "step": 545
+    },
+    {
+      "epoch": 4.788608981380066,
+      "grad_norm": 0.5573567650365066,
+      "learning_rate": 1e-06,
+      "loss": 0.10841422080993653,
+      "step": 550
+    },
+    {
+      "epoch": 4.832420591456736,
+      "grad_norm": 0.5770004286798487,
+      "learning_rate": 1e-06,
+      "loss": 0.10310661792755127,
+      "step": 555
+    },
+    {
+      "epoch": 4.876232201533407,
+      "grad_norm": 0.3568066029992377,
+      "learning_rate": 1e-06,
+      "loss": 0.10551964044570923,
+      "step": 560
+    },
+    {
+      "epoch": 4.920043811610077,
+      "grad_norm": 0.43790989026273375,
+      "learning_rate": 1e-06,
+      "loss": 0.10603113174438476,
+      "step": 565
+    },
+    {
+      "epoch": 4.9638554216867465,
+      "grad_norm": 0.5103979695681764,
+      "learning_rate": 1e-06,
+      "loss": 0.10147464275360107,
+      "step": 570
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 0.662793421930595,
+      "learning_rate": 1e-06,
+      "loss": 0.10420596599578857,
+      "step": 575
+    },
+    {
+      "epoch": 5.04381161007667,
+      "grad_norm": 0.5581404561957826,
+      "learning_rate": 1e-06,
+      "loss": 0.10490967035293579,
+      "step": 580
+    },
+    {
+      "epoch": 5.087623220153341,
+      "grad_norm": 0.4564678542607828,
+      "learning_rate": 1e-06,
+      "loss": 0.10505809783935546,
+      "step": 585
+    },
+    {
+      "epoch": 5.131434830230011,
+      "grad_norm": 0.34018141288365417,
+      "learning_rate": 1e-06,
+      "loss": 0.10050530433654785,
+      "step": 590
+    },
+    {
+      "epoch": 5.175246440306681,
+      "grad_norm": 0.37017054009516975,
+      "learning_rate": 1e-06,
+      "loss": 0.10151729583740235,
+      "step": 595
+    },
+    {
+      "epoch": 5.219058050383351,
+      "grad_norm": 0.4528758577484625,
+      "learning_rate": 1e-06,
+      "loss": 0.10294580459594727,
+      "step": 600
+    },
+    {
+      "epoch": 5.262869660460022,
+      "grad_norm": 0.5465043179143932,
+      "learning_rate": 1e-06,
+      "loss": 0.10281221866607666,
+      "step": 605
+    },
+    {
+      "epoch": 5.306681270536692,
+      "grad_norm": 0.5345857136320531,
+      "learning_rate": 1e-06,
+      "loss": 0.10342254638671874,
+      "step": 610
+    },
+    {
+      "epoch": 5.350492880613363,
+      "grad_norm": 0.32360054948918143,
+      "learning_rate": 1e-06,
+      "loss": 0.09907029271125793,
+      "step": 615
+    },
+    {
+      "epoch": 5.394304490690033,
+      "grad_norm": 0.5701646861795574,
+      "learning_rate": 1e-06,
+      "loss": 0.0970179557800293,
+      "step": 620
+    },
+    {
+      "epoch": 5.438116100766703,
+      "grad_norm": 0.4189733864525265,
+      "learning_rate": 1e-06,
+      "loss": 0.10115448236465455,
+      "step": 625
+    },
+    {
+      "epoch": 5.481927710843373,
+      "grad_norm": 0.46631011998920574,
+      "learning_rate": 1e-06,
+      "loss": 0.0997847557067871,
+      "step": 630
+    },
+    {
+      "epoch": 5.525739320920044,
+      "grad_norm": 0.48914645304576165,
+      "learning_rate": 1e-06,
+      "loss": 0.10262184143066407,
+      "step": 635
+    },
+    {
+      "epoch": 5.569550930996714,
+      "grad_norm": 0.5287684886545732,
+      "learning_rate": 1e-06,
+      "loss": 0.10598192214965821,
+      "step": 640
+    },
+    {
+      "epoch": 5.613362541073385,
+      "grad_norm": 0.4786802496887602,
+      "learning_rate": 1e-06,
+      "loss": 0.10428886413574219,
+      "step": 645
+    },
+    {
+      "epoch": 5.657174151150055,
+      "grad_norm": 0.41251457826203264,
+      "learning_rate": 1e-06,
+      "loss": 0.10103344917297363,
+      "step": 650
+    },
+    {
+      "epoch": 5.700985761226725,
+      "grad_norm": 0.3211254328605326,
+      "learning_rate": 1e-06,
+      "loss": 0.10494952201843262,
+      "step": 655
+    },
+    {
+      "epoch": 5.744797371303395,
+      "grad_norm": 0.643344448533898,
+      "learning_rate": 1e-06,
+      "loss": 0.10239223241806031,
+      "step": 660
+    },
+    {
+      "epoch": 5.788608981380066,
+      "grad_norm": 0.7493214969157797,
+      "learning_rate": 1e-06,
+      "loss": 0.10333151817321777,
+      "step": 665
+    },
+    {
+      "epoch": 5.832420591456736,
+      "grad_norm": 0.38243042285524176,
+      "learning_rate": 1e-06,
+      "loss": 0.10351332426071166,
+      "step": 670
+    },
+    {
+      "epoch": 5.876232201533407,
+      "grad_norm": 0.4141308742593925,
+      "learning_rate": 1e-06,
+      "loss": 0.10206682682037353,
+      "step": 675
+    },
+    {
+      "epoch": 5.920043811610077,
+      "grad_norm": 0.43593693884737694,
+      "learning_rate": 1e-06,
+      "loss": 0.10230462551116944,
+      "step": 680
+    },
+    {
+      "epoch": 5.9638554216867465,
+      "grad_norm": 0.702439283700418,
+      "learning_rate": 1e-06,
+      "loss": 0.10255682468414307,
+      "step": 685
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 0.7372486836303829,
+      "learning_rate": 1e-06,
+      "loss": 0.10081393718719482,
+      "step": 690
+    },
+    {
+      "epoch": 6.04381161007667,
+      "grad_norm": 0.6235268714355474,
+      "learning_rate": 1e-06,
+      "loss": 0.09922751784324646,
+      "step": 695
+    },
+    {
+      "epoch": 6.087623220153341,
+      "grad_norm": 0.7527954459719892,
+      "learning_rate": 1e-06,
+      "loss": 0.09898136854171753,
+      "step": 700
+    },
+    {
+      "epoch": 6.131434830230011,
+      "grad_norm": 0.43887725809872274,
+      "learning_rate": 1e-06,
+      "loss": 0.09697145819664002,
+      "step": 705
+    },
+    {
+      "epoch": 6.175246440306681,
+      "grad_norm": 0.46079431593402076,
+      "learning_rate": 1e-06,
+      "loss": 0.09884787201881409,
+      "step": 710
+    },
+    {
+      "epoch": 6.219058050383351,
+      "grad_norm": 0.6217947365132553,
+      "learning_rate": 1e-06,
+      "loss": 0.10021631717681885,
+      "step": 715
+    },
+    {
+      "epoch": 6.262869660460022,
+      "grad_norm": 0.40345118323171475,
+      "learning_rate": 1e-06,
+      "loss": 0.09973282217979432,
+      "step": 720
+    },
+    {
+      "epoch": 6.306681270536692,
+      "grad_norm": 0.29174428419750653,
+      "learning_rate": 1e-06,
+      "loss": 0.09687559604644776,
+      "step": 725
+    },
+    {
+      "epoch": 6.350492880613363,
+      "grad_norm": 0.5497542678258076,
+      "learning_rate": 1e-06,
+      "loss": 0.09651940464973449,
+      "step": 730
+    },
+    {
+      "epoch": 6.394304490690033,
+      "grad_norm": 0.42279950637538816,
+      "learning_rate": 1e-06,
+      "loss": 0.10130511522293091,
+      "step": 735
+    },
+    {
+      "epoch": 6.438116100766703,
+      "grad_norm": 0.3442441391853575,
+      "learning_rate": 1e-06,
+      "loss": 0.09378501176834106,
+      "step": 740
+    },
+    {
+      "epoch": 6.481927710843373,
+      "grad_norm": 0.38746248099710007,
+      "learning_rate": 1e-06,
+      "loss": 0.1019934058189392,
+      "step": 745
+    },
+    {
+      "epoch": 6.525739320920044,
+      "grad_norm": 0.3323543137951648,
+      "learning_rate": 1e-06,
+      "loss": 0.10056121349334717,
+      "step": 750
+    },
+    {
+      "epoch": 6.569550930996714,
+      "grad_norm": 0.3091945302595523,
+      "learning_rate": 1e-06,
+      "loss": 0.1020169734954834,
+      "step": 755
+    },
+    {
+      "epoch": 6.613362541073385,
+      "grad_norm": 0.47576791187520134,
+      "learning_rate": 1e-06,
+      "loss": 0.0995025634765625,
+      "step": 760
+    },
+    {
+      "epoch": 6.657174151150055,
+      "grad_norm": 0.3882847501213755,
+      "learning_rate": 1e-06,
+      "loss": 0.09973936080932617,
+      "step": 765
+    },
+    {
+      "epoch": 6.700985761226725,
+      "grad_norm": 0.4790697842646012,
+      "learning_rate": 1e-06,
+      "loss": 0.09775378704071044,
+      "step": 770
+    },
+    {
+      "epoch": 6.744797371303395,
+      "grad_norm": 0.2945721938846959,
+      "learning_rate": 1e-06,
+      "loss": 0.0982692301273346,
+      "step": 775
+    },
+    {
+      "epoch": 6.788608981380066,
+      "grad_norm": 0.5376245614002096,
+      "learning_rate": 1e-06,
+      "loss": 0.09780228734016419,
+      "step": 780
+    },
+    {
+      "epoch": 6.832420591456736,
+      "grad_norm": 0.31615139343806015,
+      "learning_rate": 1e-06,
+      "loss": 0.09756550788879395,
+      "step": 785
+    },
+    {
+      "epoch": 6.876232201533407,
+      "grad_norm": 0.5484590333810635,
+      "learning_rate": 1e-06,
+      "loss": 0.09836593866348267,
+      "step": 790
+    },
+    {
+      "epoch": 6.920043811610077,
+      "grad_norm": 0.5636458189927873,
+      "learning_rate": 1e-06,
+      "loss": 0.09832623600959778,
+      "step": 795
+    },
+    {
+      "epoch": 6.9638554216867465,
+      "grad_norm": 0.42788712278228525,
+      "learning_rate": 1e-06,
+      "loss": 0.09703341722488404,
+      "step": 800
+    },
+    {
+      "epoch": 7.0,
+      "grad_norm": 0.8392682832718135,
+      "learning_rate": 1e-06,
+      "loss": 0.10333873033523559,
+      "step": 805
+    },
+    {
+      "epoch": 7.04381161007667,
+      "grad_norm": 0.7373155101482225,
+      "learning_rate": 1e-06,
+      "loss": 0.09763166308403015,
+      "step": 810
+    },
+    {
+      "epoch": 7.087623220153341,
+      "grad_norm": 0.4842629832236003,
+      "learning_rate": 1e-06,
+      "loss": 0.09585025310516357,
+      "step": 815
+    },
+    {
+      "epoch": 7.131434830230011,
+      "grad_norm": 0.6401084148797687,
+      "learning_rate": 1e-06,
+      "loss": 0.09682692289352417,
+      "step": 820
+    },
+    {
+      "epoch": 7.175246440306681,
+      "grad_norm": 0.4629546238921947,
+      "learning_rate": 1e-06,
+      "loss": 0.09345394372940063,
+      "step": 825
+    },
+    {
+      "epoch": 7.219058050383351,
+      "grad_norm": 0.3918229484825282,
+      "learning_rate": 1e-06,
+      "loss": 0.09544733166694641,
+      "step": 830
+    },
+    {
+      "epoch": 7.262869660460022,
+      "grad_norm": 0.3796669895738244,
+      "learning_rate": 1e-06,
+      "loss": 0.09363532662391663,
+      "step": 835
+    },
+    {
+      "epoch": 7.306681270536692,
+      "grad_norm": 0.38500261411442444,
+      "learning_rate": 1e-06,
+      "loss": 0.09915844202041627,
+      "step": 840
+    },
+    {
+      "epoch": 7.350492880613363,
+      "grad_norm": 0.32402673037000806,
+      "learning_rate": 1e-06,
+      "loss": 0.08870189189910889,
+      "step": 845
+    },
+    {
+      "epoch": 7.394304490690033,
+      "grad_norm": 0.44298091288810754,
+      "learning_rate": 1e-06,
+      "loss": 0.09819216132164002,
+      "step": 850
+    },
+    {
+      "epoch": 7.438116100766703,
+      "grad_norm": 0.3497448932071372,
+      "learning_rate": 1e-06,
+      "loss": 0.10197398662567139,
+      "step": 855
+    },
+    {
+      "epoch": 7.481927710843373,
+      "grad_norm": 0.5247115892922294,
+      "learning_rate": 1e-06,
+      "loss": 0.09763568043708801,
+      "step": 860
+    },
+    {
+      "epoch": 7.525739320920044,
+      "grad_norm": 0.42546443266708117,
+      "learning_rate": 1e-06,
+      "loss": 0.0969454824924469,
+      "step": 865
+    },
+    {
+      "epoch": 7.569550930996714,
+      "grad_norm": 0.43173623974981207,
+      "learning_rate": 1e-06,
+      "loss": 0.0958714485168457,
+      "step": 870
+    },
+    {
+      "epoch": 7.613362541073385,
+      "grad_norm": 0.5433593386482033,
+      "learning_rate": 1e-06,
+      "loss": 0.09578515887260437,
+      "step": 875
+    },
+    {
+      "epoch": 7.657174151150055,
+      "grad_norm": 0.4894792516459858,
+      "learning_rate": 1e-06,
+      "loss": 0.09629914164543152,
+      "step": 880
+    },
+    {
+      "epoch": 7.700985761226725,
+      "grad_norm": 0.32699394463785353,
+      "learning_rate": 1e-06,
+      "loss": 0.09102729558944703,
+      "step": 885
+    },
+    {
+      "epoch": 7.744797371303395,
+      "grad_norm": 0.44247357247173547,
+      "learning_rate": 1e-06,
+      "loss": 0.09433148503303528,
+      "step": 890
+    },
+    {
+      "epoch": 7.788608981380066,
+      "grad_norm": 0.4493214126532042,
+      "learning_rate": 1e-06,
+      "loss": 0.09328880906105042,
+      "step": 895
+    },
+    {
+      "epoch": 7.832420591456736,
+      "grad_norm": 0.32826759250481985,
+      "learning_rate": 1e-06,
+      "loss": 0.09506406784057617,
+      "step": 900
+    },
+    {
+      "epoch": 7.876232201533407,
+      "grad_norm": 0.6557152940252813,
+      "learning_rate": 1e-06,
+      "loss": 0.09371026754379272,
+      "step": 905
+    },
+    {
+      "epoch": 7.920043811610077,
+      "grad_norm": 0.5229828944745257,
+      "learning_rate": 1e-06,
+      "loss": 0.09841654896736145,
+      "step": 910
+    },
+    {
+      "epoch": 7.9638554216867465,
+      "grad_norm": 0.40411088680364343,
+      "learning_rate": 1e-06,
+      "loss": 0.09659086465835572,
+      "step": 915
+    },
+    {
+      "epoch": 8.0,
+      "grad_norm": 0.7796862766907513,
+      "learning_rate": 1e-06,
+      "loss": 0.09889492392539978,
+      "step": 920
+    },
+    {
+      "epoch": 8.04381161007667,
+      "grad_norm": 0.6980514755693372,
+      "learning_rate": 1e-06,
+      "loss": 0.09768669605255127,
+      "step": 925
+    },
+    {
+      "epoch": 8.08762322015334,
+      "grad_norm": 0.3384881538110318,
+      "learning_rate": 1e-06,
+      "loss": 0.0911449134349823,
+      "step": 930
+    },
+    {
+      "epoch": 8.131434830230011,
+      "grad_norm": 0.36321989445747427,
+      "learning_rate": 1e-06,
+      "loss": 0.0907370924949646,
+      "step": 935
+    },
+    {
+      "epoch": 8.175246440306681,
+      "grad_norm": 0.299540201934078,
+      "learning_rate": 1e-06,
+      "loss": 0.09337329864501953,
+      "step": 940
+    },
+    {
+      "epoch": 8.219058050383351,
+      "grad_norm": 0.3851643395810515,
+      "learning_rate": 1e-06,
+      "loss": 0.09015904664993286,
+      "step": 945
+    },
+    {
+      "epoch": 8.262869660460021,
+      "grad_norm": 0.3392377456783416,
+      "learning_rate": 1e-06,
+      "loss": 0.09023426771163941,
+      "step": 950
+    },
+    {
+      "epoch": 8.306681270536693,
+      "grad_norm": 0.5203920708319888,
+      "learning_rate": 1e-06,
+      "loss": 0.09258170127868652,
+      "step": 955
+    },
+    {
+      "epoch": 8.350492880613363,
+      "grad_norm": 0.6828437820745489,
+      "learning_rate": 1e-06,
+      "loss": 0.0952469289302826,
+      "step": 960
+    },
+    {
+      "epoch": 8.394304490690033,
+      "grad_norm": 0.5580576421664939,
+      "learning_rate": 1e-06,
+      "loss": 0.09448376893997193,
+      "step": 965
+    },
+    {
+      "epoch": 8.438116100766702,
+      "grad_norm": 1.0083910547301178,
+      "learning_rate": 1e-06,
+      "loss": 0.08862833976745606,
+      "step": 970
+    },
+    {
+      "epoch": 8.481927710843374,
+      "grad_norm": 0.5465948630695099,
+      "learning_rate": 1e-06,
+      "loss": 0.09315670728683471,
+      "step": 975
+    },
+    {
+      "epoch": 8.525739320920044,
+      "grad_norm": 0.5186394479206893,
+      "learning_rate": 1e-06,
+      "loss": 0.09288825988769531,
+      "step": 980
+    },
+    {
+      "epoch": 8.569550930996714,
+      "grad_norm": 0.6601025967746997,
+      "learning_rate": 1e-06,
+      "loss": 0.08876540064811707,
+      "step": 985
+    },
+    {
+      "epoch": 8.613362541073384,
+      "grad_norm": 0.602596867944175,
+      "learning_rate": 1e-06,
+      "loss": 0.09005357027053833,
+      "step": 990
+    },
+    {
+      "epoch": 8.657174151150056,
+      "grad_norm": 0.5420986848870678,
+      "learning_rate": 1e-06,
+      "loss": 0.0924445629119873,
+      "step": 995
+    },
+    {
+      "epoch": 8.700985761226725,
+      "grad_norm": 0.5418997342034354,
+      "learning_rate": 1e-06,
+      "loss": 0.09406256079673767,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 4280,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 38,
+  "save_steps": 200,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4782378197188608.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56c4182f80e2a2aedaa1a9daa7033aeb41814d2e7dcc1ac6e0894f0842545a53
+size 7761

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 ZERO_STAGE not 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("Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info("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)