Upload checkpoint-3000 from E118

checkpoints/E118/checkpoint-3000/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|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
+}

checkpoints/E118/checkpoint-3000/config.json

@@ -0,0 +1,84 @@
+{
+  "architectures": [
+    "Qwen2_5_VLForConditionalGenerationWithVGGT"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "feature_fusion_method": "deepstack_language_add",
+  "fusion_num_layers": 1,
+  "geometry_encoder_layers": [
+    11,
+    17,
+    22
+  ],
+  "geometry_encoder_type": "vggt",
+  "geometry_fusion_layers": [
+    0,
+    1,
+    2
+  ],
+  "geometry_merger_type": "mlp",
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "image_token_id": 151655,
+  "include_camera_token": false,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 128000,
+  "max_window_layers": 70,
+  "model_type": "qwen2_5_vl",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 36,
+  "num_key_value_heads": 2,
+  "pos_encoding_type": "none",
+  "reference_frame": "first",
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": {
+    "mrope_section": [
+      16,
+      24,
+      24
+    ],
+    "rope_type": "default",
+    "type": "default"
+  },
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.50.0",
+  "use_cache": false,
+  "use_geometry_encoder": true,
+  "use_sliding_window": false,
+  "video_token_id": 151656,
+  "vision_config": {
+    "depth": 32,
+    "fullatt_block_indexes": [
+      7,
+      15,
+      23,
+      31
+    ],
+    "hidden_act": "silu",
+    "hidden_size": 1280,
+    "in_channels": 3,
+    "in_chans": 3,
+    "intermediate_size": 3420,
+    "model_type": "qwen2_5_vl",
+    "num_heads": 16,
+    "out_hidden_size": 2048,
+    "patch_size": 14,
+    "spatial_merge_size": 2,
+    "spatial_patch_size": 14,
+    "temporal_patch_size": 2,
+    "tokens_per_second": 2,
+    "torch_dtype": "bfloat16",
+    "window_size": 112
+  },
+  "vision_end_token_id": 151653,
+  "vision_language_fusion_layers": null,
+  "vision_start_token_id": 151652,
+  "vision_token_id": 151654,
+  "vocab_size": 151936
+}

checkpoints/E118/checkpoint-3000/generation_config.json

@@ -0,0 +1,12 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.05,
+  "temperature": 1e-06,
+  "transformers_version": "4.50.0"
+}

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6c22ae910e4b0b6cc438e4c01470778bbd0bdb94afb4bb4876e6c0c97c4fbb9
+size 4817695729

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d1062eb4317c0025bc065d821f67b40e25cf0f84c0aa83fcad06c7507a4689a
+size 4817698545

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:066a786debce803eee79999b2a97ae650de7103df63827858c65f477b7d5557b
+size 4817699121

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36029d966cfe7b004ccbc097a2f369a1310771acbf3ef762682881f22d8d0c89
+size 4817699569

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e43eeb848c864bb648493fc5e8fd721b3060b0c724560b42db120ea8614d9ba3
+size 4817698993

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1197119d968bc9ac8b84cea9b4b9f4602c7a76820635ab29ca52145c878265fd
+size 4817699249

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68837240ab45ed08a069d9e51766945e9b7a78259342e1f98135b3f06834c8f6
+size 4817699057

checkpoints/E118/checkpoint-3000/global_step3000/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e5f43400dc31e88e4924145149bc3d86fd6f8b39a6d39bbb5c1e4ed48a63122
+size 4817699377

checkpoints/E118/checkpoint-3000/global_step3000/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e9d220b00635d474c12dcbb6a66c7ce9aa963d6fca8855aa79a1a6d7ae2ed62
+size 12736254037

checkpoints/E118/checkpoint-3000/latest

@@ -0,0 +1 @@
+global_step3000

checkpoints/E118/checkpoint-3000/model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:325e9b29dde61fe900c9b5d7e516ca8ef768244a8a64407e2ace85f4098586b2
+size 4962140952

checkpoints/E118/checkpoint-3000/model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95d50bf28221f676654121ff676f110bb76c288556c0faa7bf88c16ba30e754b
+size 4869006472

checkpoints/E118/checkpoint-3000/model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f7a27b94b46303c1e36e7c16a7ad31b4ba9b63039f2bad5302b9ebd13ed7289
+size 622329984

checkpoints/E118/checkpoint-3000/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93366b78442d88be019e36f3ea689d21ed498b09e95d530b448530a143cc3f36
+size 16325

checkpoints/E118/checkpoint-3000/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d721cb58d3bd65db773d5c1f43927f5aad0aeb38d44f1201f9142e93b467b6ed
+size 16325

checkpoints/E118/checkpoint-3000/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8cd3dabf2f4ae849586a5b78bf256d3ee47426368adf39ef0cfa60f97e31328
+size 16325

checkpoints/E118/checkpoint-3000/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a6d3c132b137b6645ecb4941c37da51f062228a8f875e480ad1ab2786df8a46
+size 16325

checkpoints/E118/checkpoint-3000/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70602ef413ff4a3a2b3c27c82c937c3edccab869afd9899600831facd1efa37d
+size 16325

checkpoints/E118/checkpoint-3000/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5da8055d6f4011e5ec500f5e23cc87cbfed3e0e1aac3d8d331d42698da8e6aa6
+size 16325

checkpoints/E118/checkpoint-3000/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3eca2914b1b5c55d14aafdc6a7517c617a86a987344214e557c3c5b9d74bc10
+size 16325

checkpoints/E118/checkpoint-3000/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26f3977ddd6e391a32d3384758177f0212c6b30f62092335828dc6f5a279396f
+size 16325

checkpoints/E118/checkpoint-3000/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ab16fb6d9408aabd351dacf499a039b866785c0ccf0c945247fbe9725001837
+size 1465

checkpoints/E118/checkpoint-3000/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
+  }
+}

checkpoints/E118/checkpoint-3000/tokenizer_config.json

@@ -0,0 +1,209 @@
+{
+  "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
+    }
+  },
+  "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": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 12800,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoints/E118/checkpoint-3000/trainer_state.json

@@ -0,0 +1,2134 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9065498224673264,
+  "eval_steps": 500,
+  "global_step": 3000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0030218327415577548,
+      "grad_norm": 164.14991760253906,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 13.8259,
+      "step": 10
+    },
+    {
+      "epoch": 0.0060436654831155096,
+      "grad_norm": 79.00243377685547,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 13.0608,
+      "step": 20
+    },
+    {
+      "epoch": 0.009065498224673264,
+      "grad_norm": 51.88568878173828,
+      "learning_rate": 3e-06,
+      "loss": 12.0917,
+      "step": 30
+    },
+    {
+      "epoch": 0.012087330966231019,
+      "grad_norm": 48.75140380859375,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 10.3111,
+      "step": 40
+    },
+    {
+      "epoch": 0.015109163707788774,
+      "grad_norm": 44.99660873413086,
+      "learning_rate": 5e-06,
+      "loss": 9.553,
+      "step": 50
+    },
+    {
+      "epoch": 0.01813099644934653,
+      "grad_norm": 41.01961135864258,
+      "learning_rate": 6e-06,
+      "loss": 8.3421,
+      "step": 60
+    },
+    {
+      "epoch": 0.021152829190904283,
+      "grad_norm": 31.11094856262207,
+      "learning_rate": 7e-06,
+      "loss": 6.9212,
+      "step": 70
+    },
+    {
+      "epoch": 0.024174661932462038,
+      "grad_norm": 51.63551712036133,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 6.7758,
+      "step": 80
+    },
+    {
+      "epoch": 0.027196494674019793,
+      "grad_norm": 30.700332641601562,
+      "learning_rate": 9e-06,
+      "loss": 7.0498,
+      "step": 90
+    },
+    {
+      "epoch": 0.030218327415577548,
+      "grad_norm": 37.647708892822266,
+      "learning_rate": 1e-05,
+      "loss": 6.7113,
+      "step": 100
+    },
+    {
+      "epoch": 0.0332401601571353,
+      "grad_norm": 37.32933807373047,
+      "learning_rate": 9.999760394462267e-06,
+      "loss": 7.3592,
+      "step": 110
+    },
+    {
+      "epoch": 0.03626199289869306,
+      "grad_norm": 38.281829833984375,
+      "learning_rate": 9.999041600813393e-06,
+      "loss": 7.0046,
+      "step": 120
+    },
+    {
+      "epoch": 0.03928382564025081,
+      "grad_norm": 31.066274642944336,
+      "learning_rate": 9.997843687944153e-06,
+      "loss": 6.2365,
+      "step": 130
+    },
+    {
+      "epoch": 0.04230565838180857,
+      "grad_norm": 29.11764144897461,
+      "learning_rate": 9.996166770665168e-06,
+      "loss": 5.579,
+      "step": 140
+    },
+    {
+      "epoch": 0.04532749112336632,
+      "grad_norm": 33.222808837890625,
+      "learning_rate": 9.994011009695908e-06,
+      "loss": 5.6607,
+      "step": 150
+    },
+    {
+      "epoch": 0.048349323864924076,
+      "grad_norm": 31.273977279663086,
+      "learning_rate": 9.991376611649278e-06,
+      "loss": 6.0873,
+      "step": 160
+    },
+    {
+      "epoch": 0.05137115660648183,
+      "grad_norm": 27.030546188354492,
+      "learning_rate": 9.988263829011821e-06,
+      "loss": 5.4012,
+      "step": 170
+    },
+    {
+      "epoch": 0.054392989348039586,
+      "grad_norm": 25.91176986694336,
+      "learning_rate": 9.984672960119523e-06,
+      "loss": 5.3626,
+      "step": 180
+    },
+    {
+      "epoch": 0.05741482208959734,
+      "grad_norm": 39.797523498535156,
+      "learning_rate": 9.980604349129212e-06,
+      "loss": 5.7565,
+      "step": 190
+    },
+    {
+      "epoch": 0.060436654831155096,
+      "grad_norm": 26.1552734375,
+      "learning_rate": 9.976058385985575e-06,
+      "loss": 5.1983,
+      "step": 200
+    },
+    {
+      "epoch": 0.06345848757271286,
+      "grad_norm": 34.96975326538086,
+      "learning_rate": 9.971035506383791e-06,
+      "loss": 5.6103,
+      "step": 210
+    },
+    {
+      "epoch": 0.0664803203142706,
+      "grad_norm": 30.275611877441406,
+      "learning_rate": 9.96553619172777e-06,
+      "loss": 5.082,
+      "step": 220
+    },
+    {
+      "epoch": 0.06950215305582837,
+      "grad_norm": 31.659500122070312,
+      "learning_rate": 9.959560969084004e-06,
+      "loss": 5.335,
+      "step": 230
+    },
+    {
+      "epoch": 0.07252398579738611,
+      "grad_norm": 25.909767150878906,
+      "learning_rate": 9.953110411131073e-06,
+      "loss": 4.7812,
+      "step": 240
+    },
+    {
+      "epoch": 0.07554581853894388,
+      "grad_norm": 29.12919044494629,
+      "learning_rate": 9.946185136104736e-06,
+      "loss": 5.437,
+      "step": 250
+    },
+    {
+      "epoch": 0.07856765128050162,
+      "grad_norm": 29.44923973083496,
+      "learning_rate": 9.938785807738692e-06,
+      "loss": 4.7933,
+      "step": 260
+    },
+    {
+      "epoch": 0.08158948402205939,
+      "grad_norm": 30.55126953125,
+      "learning_rate": 9.930913135200964e-06,
+      "loss": 5.403,
+      "step": 270
+    },
+    {
+      "epoch": 0.08461131676361713,
+      "grad_norm": 23.430017471313477,
+      "learning_rate": 9.922567873025924e-06,
+      "loss": 5.2349,
+      "step": 280
+    },
+    {
+      "epoch": 0.0876331495051749,
+      "grad_norm": 29.15253448486328,
+      "learning_rate": 9.913750821041988e-06,
+      "loss": 4.5357,
+      "step": 290
+    },
+    {
+      "epoch": 0.09065498224673264,
+      "grad_norm": 24.52574920654297,
+      "learning_rate": 9.904462824294945e-06,
+      "loss": 4.5873,
+      "step": 300
+    },
+    {
+      "epoch": 0.0936768149882904,
+      "grad_norm": 24.67723846435547,
+      "learning_rate": 9.894704772966978e-06,
+      "loss": 4.474,
+      "step": 310
+    },
+    {
+      "epoch": 0.09669864772984815,
+      "grad_norm": 24.382055282592773,
+      "learning_rate": 9.884477602291343e-06,
+      "loss": 4.5008,
+      "step": 320
+    },
+    {
+      "epoch": 0.09972048047140591,
+      "grad_norm": 30.42603302001953,
+      "learning_rate": 9.873782292462727e-06,
+      "loss": 4.3567,
+      "step": 330
+    },
+    {
+      "epoch": 0.10274231321296366,
+      "grad_norm": 26.170686721801758,
+      "learning_rate": 9.862619868543323e-06,
+      "loss": 8.1762,
+      "step": 340
+    },
+    {
+      "epoch": 0.10576414595452142,
+      "grad_norm": 38.36408996582031,
+      "learning_rate": 9.850991400364557e-06,
+      "loss": 5.1626,
+      "step": 350
+    },
+    {
+      "epoch": 0.10878597869607917,
+      "grad_norm": 25.83894157409668,
+      "learning_rate": 9.838898002424586e-06,
+      "loss": 6.0757,
+      "step": 360
+    },
+    {
+      "epoch": 0.11180781143763693,
+      "grad_norm": 29.732688903808594,
+      "learning_rate": 9.826340833781448e-06,
+      "loss": 5.8358,
+      "step": 370
+    },
+    {
+      "epoch": 0.11482964417919468,
+      "grad_norm": 18.970821380615234,
+      "learning_rate": 9.813321097942005e-06,
+      "loss": 5.1022,
+      "step": 380
+    },
+    {
+      "epoch": 0.11785147692075244,
+      "grad_norm": 27.406240463256836,
+      "learning_rate": 9.79984004274658e-06,
+      "loss": 4.8694,
+      "step": 390
+    },
+    {
+      "epoch": 0.12087330966231019,
+      "grad_norm": 22.173940658569336,
+      "learning_rate": 9.785898960249365e-06,
+      "loss": 4.2529,
+      "step": 400
+    },
+    {
+      "epoch": 0.12389514240386795,
+      "grad_norm": 19.587993621826172,
+      "learning_rate": 9.771499186594586e-06,
+      "loss": 5.0707,
+      "step": 410
+    },
+    {
+      "epoch": 0.12691697514542571,
+      "grad_norm": 26.870187759399414,
+      "learning_rate": 9.756642101888449e-06,
+      "loss": 5.1808,
+      "step": 420
+    },
+    {
+      "epoch": 0.12993880788698345,
+      "grad_norm": 24.28529167175293,
+      "learning_rate": 9.74132913006686e-06,
+      "loss": 3.4672,
+      "step": 430
+    },
+    {
+      "epoch": 0.1329606406285412,
+      "grad_norm": 20.661434173583984,
+      "learning_rate": 9.725561738758956e-06,
+      "loss": 3.3511,
+      "step": 440
+    },
+    {
+      "epoch": 0.13598247337009897,
+      "grad_norm": 22.3953914642334,
+      "learning_rate": 9.709341439146452e-06,
+      "loss": 5.1124,
+      "step": 450
+    },
+    {
+      "epoch": 0.13900430611165673,
+      "grad_norm": 27.169279098510742,
+      "learning_rate": 9.692669785818787e-06,
+      "loss": 6.4226,
+      "step": 460
+    },
+    {
+      "epoch": 0.14202613885321447,
+      "grad_norm": 34.868682861328125,
+      "learning_rate": 9.675548376624149e-06,
+      "loss": 5.7534,
+      "step": 470
+    },
+    {
+      "epoch": 0.14504797159477223,
+      "grad_norm": 27.703248977661133,
+      "learning_rate": 9.657978852516318e-06,
+      "loss": 5.6828,
+      "step": 480
+    },
+    {
+      "epoch": 0.14806980433633,
+      "grad_norm": 21.426271438598633,
+      "learning_rate": 9.639962897397405e-06,
+      "loss": 4.1847,
+      "step": 490
+    },
+    {
+      "epoch": 0.15109163707788775,
+      "grad_norm": 21.853469848632812,
+      "learning_rate": 9.621502237956452e-06,
+      "loss": 5.0356,
+      "step": 500
+    },
+    {
+      "epoch": 0.1541134698194455,
+      "grad_norm": 19.003141403198242,
+      "learning_rate": 9.602598643503957e-06,
+      "loss": 3.2694,
+      "step": 510
+    },
+    {
+      "epoch": 0.15713530256100325,
+      "grad_norm": 28.9671688079834,
+      "learning_rate": 9.583253925802283e-06,
+      "loss": 4.2167,
+      "step": 520
+    },
+    {
+      "epoch": 0.160157135302561,
+      "grad_norm": 22.439605712890625,
+      "learning_rate": 9.563469938892023e-06,
+      "loss": 4.8531,
+      "step": 530
+    },
+    {
+      "epoch": 0.16317896804411877,
+      "grad_norm": 23.49073028564453,
+      "learning_rate": 9.543248578914309e-06,
+      "loss": 3.222,
+      "step": 540
+    },
+    {
+      "epoch": 0.1662008007856765,
+      "grad_norm": 23.804365158081055,
+      "learning_rate": 9.522591783929069e-06,
+      "loss": 4.829,
+      "step": 550
+    },
+    {
+      "epoch": 0.16922263352723427,
+      "grad_norm": 19.732295989990234,
+      "learning_rate": 9.501501533729297e-06,
+      "loss": 4.128,
+      "step": 560
+    },
+    {
+      "epoch": 0.17224446626879203,
+      "grad_norm": 33.48932647705078,
+      "learning_rate": 9.479979849651287e-06,
+      "loss": 5.7656,
+      "step": 570
+    },
+    {
+      "epoch": 0.1752662990103498,
+      "grad_norm": 24.05331039428711,
+      "learning_rate": 9.45802879438091e-06,
+      "loss": 6.3455,
+      "step": 580
+    },
+    {
+      "epoch": 0.17828813175190752,
+      "grad_norm": 23.51070213317871,
+      "learning_rate": 9.43565047175593e-06,
+      "loss": 4.1453,
+      "step": 590
+    },
+    {
+      "epoch": 0.1813099644934653,
+      "grad_norm": 25.37734603881836,
+      "learning_rate": 9.412847026564359e-06,
+      "loss": 3.9563,
+      "step": 600
+    },
+    {
+      "epoch": 0.18433179723502305,
+      "grad_norm": 21.561195373535156,
+      "learning_rate": 9.389620644338893e-06,
+      "loss": 4.1523,
+      "step": 610
+    },
+    {
+      "epoch": 0.1873536299765808,
+      "grad_norm": 35.891597747802734,
+      "learning_rate": 9.365973551147453e-06,
+      "loss": 4.6762,
+      "step": 620
+    },
+    {
+      "epoch": 0.19037546271813854,
+      "grad_norm": 22.53954315185547,
+      "learning_rate": 9.341908013379832e-06,
+      "loss": 4.7275,
+      "step": 630
+    },
+    {
+      "epoch": 0.1933972954596963,
+      "grad_norm": 25.00588607788086,
+      "learning_rate": 9.317426337530477e-06,
+      "loss": 4.0427,
+      "step": 640
+    },
+    {
+      "epoch": 0.19641912820125407,
+      "grad_norm": 26.281129837036133,
+      "learning_rate": 9.292530869977432e-06,
+      "loss": 5.5388,
+      "step": 650
+    },
+    {
+      "epoch": 0.19944096094281183,
+      "grad_norm": 23.323177337646484,
+      "learning_rate": 9.26722399675745e-06,
+      "loss": 3.1677,
+      "step": 660
+    },
+    {
+      "epoch": 0.20246279368436956,
+      "grad_norm": 21.66280174255371,
+      "learning_rate": 9.24150814333732e-06,
+      "loss": 3.9482,
+      "step": 670
+    },
+    {
+      "epoch": 0.20548462642592732,
+      "grad_norm": 19.23092269897461,
+      "learning_rate": 9.215385774381395e-06,
+      "loss": 6.2152,
+      "step": 680
+    },
+    {
+      "epoch": 0.2085064591674851,
+      "grad_norm": 41.06984329223633,
+      "learning_rate": 9.188859393515382e-06,
+      "loss": 4.8277,
+      "step": 690
+    },
+    {
+      "epoch": 0.21152829190904285,
+      "grad_norm": 22.713626861572266,
+      "learning_rate": 9.16193154308638e-06,
+      "loss": 6.0234,
+      "step": 700
+    },
+    {
+      "epoch": 0.21455012465060058,
+      "grad_norm": 34.92478942871094,
+      "learning_rate": 9.13460480391922e-06,
+      "loss": 6.2043,
+      "step": 710
+    },
+    {
+      "epoch": 0.21757195739215834,
+      "grad_norm": 37.5953254699707,
+      "learning_rate": 9.106881795069116e-06,
+      "loss": 6.4758,
+      "step": 720
+    },
+    {
+      "epoch": 0.2205937901337161,
+      "grad_norm": 19.387893676757812,
+      "learning_rate": 9.078765173570649e-06,
+      "loss": 3.1607,
+      "step": 730
+    },
+    {
+      "epoch": 0.22361562287527387,
+      "grad_norm": 19.037372589111328,
+      "learning_rate": 9.0502576341831e-06,
+      "loss": 4.0532,
+      "step": 740
+    },
+    {
+      "epoch": 0.2266374556168316,
+      "grad_norm": 20.597227096557617,
+      "learning_rate": 9.02136190913219e-06,
+      "loss": 5.4351,
+      "step": 750
+    },
+    {
+      "epoch": 0.22965928835838936,
+      "grad_norm": 37.33137130737305,
+      "learning_rate": 8.99208076784822e-06,
+      "loss": 4.6044,
+      "step": 760
+    },
+    {
+      "epoch": 0.23268112109994712,
+      "grad_norm": 16.10256004333496,
+      "learning_rate": 8.962417016700624e-06,
+      "loss": 3.0566,
+      "step": 770
+    },
+    {
+      "epoch": 0.2357029538415049,
+      "grad_norm": 23.969058990478516,
+      "learning_rate": 8.932373498729026e-06,
+      "loss": 4.635,
+      "step": 780
+    },
+    {
+      "epoch": 0.23872478658306262,
+      "grad_norm": 24.59278106689453,
+      "learning_rate": 8.901953093370734e-06,
+      "loss": 3.9973,
+      "step": 790
+    },
+    {
+      "epoch": 0.24174661932462038,
+      "grad_norm": 25.19046401977539,
+      "learning_rate": 8.871158716184784e-06,
+      "loss": 3.9002,
+      "step": 800
+    },
+    {
+      "epoch": 0.24476845206617814,
+      "grad_norm": 32.421531677246094,
+      "learning_rate": 8.839993318572497e-06,
+      "loss": 4.8339,
+      "step": 810
+    },
+    {
+      "epoch": 0.2477902848077359,
+      "grad_norm": 18.438983917236328,
+      "learning_rate": 8.808459887494617e-06,
+      "loss": 3.0604,
+      "step": 820
+    },
+    {
+      "epoch": 0.25081211754929367,
+      "grad_norm": 18.770906448364258,
+      "learning_rate": 8.77656144518502e-06,
+      "loss": 3.8172,
+      "step": 830
+    },
+    {
+      "epoch": 0.25383395029085143,
+      "grad_norm": 17.892440795898438,
+      "learning_rate": 8.744301048861083e-06,
+      "loss": 2.9072,
+      "step": 840
+    },
+    {
+      "epoch": 0.25685578303240914,
+      "grad_norm": 22.90224838256836,
+      "learning_rate": 8.711681790430646e-06,
+      "loss": 3.0475,
+      "step": 850
+    },
+    {
+      "epoch": 0.2598776157739669,
+      "grad_norm": 25.596424102783203,
+      "learning_rate": 8.678706796195694e-06,
+      "loss": 4.7575,
+      "step": 860
+    },
+    {
+      "epoch": 0.26289944851552466,
+      "grad_norm": 22.27895736694336,
+      "learning_rate": 8.645379226552712e-06,
+      "loss": 3.7133,
+      "step": 870
+    },
+    {
+      "epoch": 0.2659212812570824,
+      "grad_norm": 20.32448387145996,
+      "learning_rate": 8.611702275689805e-06,
+      "loss": 4.6497,
+      "step": 880
+    },
+    {
+      "epoch": 0.2689431139986402,
+      "grad_norm": 34.19053649902344,
+      "learning_rate": 8.577679171280538e-06,
+      "loss": 4.4895,
+      "step": 890
+    },
+    {
+      "epoch": 0.27196494674019794,
+      "grad_norm": 18.900352478027344,
+      "learning_rate": 8.543313174174601e-06,
+      "loss": 5.2694,
+      "step": 900
+    },
+    {
+      "epoch": 0.2749867794817557,
+      "grad_norm": 22.090261459350586,
+      "learning_rate": 8.508607578085281e-06,
+      "loss": 3.8479,
+      "step": 910
+    },
+    {
+      "epoch": 0.27800861222331347,
+      "grad_norm": 20.50776481628418,
+      "learning_rate": 8.473565709273786e-06,
+      "loss": 3.837,
+      "step": 920
+    },
+    {
+      "epoch": 0.2810304449648712,
+      "grad_norm": 17.40482521057129,
+      "learning_rate": 8.438190926230439e-06,
+      "loss": 3.8182,
+      "step": 930
+    },
+    {
+      "epoch": 0.28405227770642894,
+      "grad_norm": 32.253265380859375,
+      "learning_rate": 8.40248661935281e-06,
+      "loss": 4.3698,
+      "step": 940
+    },
+    {
+      "epoch": 0.2870741104479867,
+      "grad_norm": 22.86214256286621,
+      "learning_rate": 8.366456210620756e-06,
+      "loss": 3.0836,
+      "step": 950
+    },
+    {
+      "epoch": 0.29009594318954446,
+      "grad_norm": 29.654722213745117,
+      "learning_rate": 8.330103153268464e-06,
+      "loss": 3.7893,
+      "step": 960
+    },
+    {
+      "epoch": 0.2931177759311022,
+      "grad_norm": 22.348018646240234,
+      "learning_rate": 8.29343093145347e-06,
+      "loss": 3.7182,
+      "step": 970
+    },
+    {
+      "epoch": 0.29613960867266,
+      "grad_norm": 21.902996063232422,
+      "learning_rate": 8.25644305992275e-06,
+      "loss": 3.6397,
+      "step": 980
+    },
+    {
+      "epoch": 0.29916144141421774,
+      "grad_norm": 18.903818130493164,
+      "learning_rate": 8.21914308367584e-06,
+      "loss": 4.5713,
+      "step": 990
+    },
+    {
+      "epoch": 0.3021832741557755,
+      "grad_norm": 20.676620483398438,
+      "learning_rate": 8.181534577625088e-06,
+      "loss": 3.7566,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3052051068973332,
+      "grad_norm": 13.126559257507324,
+      "learning_rate": 8.143621146253022e-06,
+      "loss": 4.6186,
+      "step": 1010
+    },
+    {
+      "epoch": 0.308226939638891,
+      "grad_norm": 16.96494483947754,
+      "learning_rate": 8.105406423266884e-06,
+      "loss": 4.6396,
+      "step": 1020
+    },
+    {
+      "epoch": 0.31124877238044873,
+      "grad_norm": 18.15988540649414,
+      "learning_rate": 8.066894071250374e-06,
+      "loss": 4.4521,
+      "step": 1030
+    },
+    {
+      "epoch": 0.3142706051220065,
+      "grad_norm": 18.560087203979492,
+      "learning_rate": 8.02808778131262e-06,
+      "loss": 3.7732,
+      "step": 1040
+    },
+    {
+      "epoch": 0.31729243786356426,
+      "grad_norm": 16.505374908447266,
+      "learning_rate": 7.988991272734407e-06,
+      "loss": 4.4711,
+      "step": 1050
+    },
+    {
+      "epoch": 0.320314270605122,
+      "grad_norm": 20.265331268310547,
+      "learning_rate": 7.94960829261172e-06,
+      "loss": 4.4707,
+      "step": 1060
+    },
+    {
+      "epoch": 0.3233361033466798,
+      "grad_norm": 19.918270111083984,
+      "learning_rate": 7.909942615496613e-06,
+      "loss": 4.6998,
+      "step": 1070
+    },
+    {
+      "epoch": 0.32635793608823754,
+      "grad_norm": 29.365598678588867,
+      "learning_rate": 7.869998043035442e-06,
+      "loss": 5.3852,
+      "step": 1080
+    },
+    {
+      "epoch": 0.32937976882979525,
+      "grad_norm": 15.32125186920166,
+      "learning_rate": 7.829778403604512e-06,
+      "loss": 5.0378,
+      "step": 1090
+    },
+    {
+      "epoch": 0.332401601571353,
+      "grad_norm": 33.20755386352539,
+      "learning_rate": 7.789287551943158e-06,
+      "loss": 6.0804,
+      "step": 1100
+    },
+    {
+      "epoch": 0.3354234343129108,
+      "grad_norm": 17.992769241333008,
+      "learning_rate": 7.748529368784293e-06,
+      "loss": 4.541,
+      "step": 1110
+    },
+    {
+      "epoch": 0.33844526705446853,
+      "grad_norm": 17.076702117919922,
+      "learning_rate": 7.707507760482473e-06,
+      "loss": 6.1712,
+      "step": 1120
+    },
+    {
+      "epoch": 0.3414670997960263,
+      "grad_norm": 16.922510147094727,
+      "learning_rate": 7.666226658639507e-06,
+      "loss": 3.763,
+      "step": 1130
+    },
+    {
+      "epoch": 0.34448893253758406,
+      "grad_norm": 23.991716384887695,
+      "learning_rate": 7.624690019727636e-06,
+      "loss": 3.6615,
+      "step": 1140
+    },
+    {
+      "epoch": 0.3475107652791418,
+      "grad_norm": 21.76751708984375,
+      "learning_rate": 7.58290182471034e-06,
+      "loss": 4.5626,
+      "step": 1150
+    },
+    {
+      "epoch": 0.3505325980206996,
+      "grad_norm": 15.035407066345215,
+      "learning_rate": 7.5408660786607976e-06,
+      "loss": 3.6885,
+      "step": 1160
+    },
+    {
+      "epoch": 0.3535544307622573,
+      "grad_norm": 22.35795021057129,
+      "learning_rate": 7.498586810378019e-06,
+      "loss": 2.9841,
+      "step": 1170
+    },
+    {
+      "epoch": 0.35657626350381505,
+      "grad_norm": 19.678627014160156,
+      "learning_rate": 7.456068072000731e-06,
+      "loss": 2.7928,
+      "step": 1180
+    },
+    {
+      "epoch": 0.3595980962453728,
+      "grad_norm": 22.308584213256836,
+      "learning_rate": 7.4133139386190026e-06,
+      "loss": 4.5458,
+      "step": 1190
+    },
+    {
+      "epoch": 0.3626199289869306,
+      "grad_norm": 18.34815216064453,
+      "learning_rate": 7.3703285078836796e-06,
+      "loss": 5.2143,
+      "step": 1200
+    },
+    {
+      "epoch": 0.36564176172848833,
+      "grad_norm": 16.58567237854004,
+      "learning_rate": 7.3271158996136625e-06,
+      "loss": 3.7281,
+      "step": 1210
+    },
+    {
+      "epoch": 0.3686635944700461,
+      "grad_norm": 21.284473419189453,
+      "learning_rate": 7.283680255401049e-06,
+      "loss": 4.4133,
+      "step": 1220
+    },
+    {
+      "epoch": 0.37168542721160386,
+      "grad_norm": 17.73834800720215,
+      "learning_rate": 7.240025738214193e-06,
+      "loss": 6.2388,
+      "step": 1230
+    },
+    {
+      "epoch": 0.3747072599531616,
+      "grad_norm": 31.61699867248535,
+      "learning_rate": 7.196156531998718e-06,
+      "loss": 4.5308,
+      "step": 1240
+    },
+    {
+      "epoch": 0.3777290926947193,
+      "grad_norm": 19.153860092163086,
+      "learning_rate": 7.152076841276527e-06,
+      "loss": 3.6474,
+      "step": 1250
+    },
+    {
+      "epoch": 0.3807509254362771,
+      "grad_norm": 19.91427993774414,
+      "learning_rate": 7.1077908907428154e-06,
+      "loss": 3.7991,
+      "step": 1260
+    },
+    {
+      "epoch": 0.38377275817783485,
+      "grad_norm": 31.496000289916992,
+      "learning_rate": 7.063302924861182e-06,
+      "loss": 3.8615,
+      "step": 1270
+    },
+    {
+      "epoch": 0.3867945909193926,
+      "grad_norm": 21.705524444580078,
+      "learning_rate": 7.018617207456821e-06,
+      "loss": 3.5938,
+      "step": 1280
+    },
+    {
+      "epoch": 0.3898164236609504,
+      "grad_norm": 18.876136779785156,
+      "learning_rate": 6.973738021307872e-06,
+      "loss": 3.6873,
+      "step": 1290
+    },
+    {
+      "epoch": 0.39283825640250813,
+      "grad_norm": 18.49332046508789,
+      "learning_rate": 6.9286696677349455e-06,
+      "loss": 5.946,
+      "step": 1300
+    },
+    {
+      "epoch": 0.3958600891440659,
+      "grad_norm": 15.16926383972168,
+      "learning_rate": 6.883416466188881e-06,
+      "loss": 3.6912,
+      "step": 1310
+    },
+    {
+      "epoch": 0.39888192188562366,
+      "grad_norm": 15.419941902160645,
+      "learning_rate": 6.837982753836755e-06,
+      "loss": 2.8732,
+      "step": 1320
+    },
+    {
+      "epoch": 0.40190375462718136,
+      "grad_norm": 37.54107666015625,
+      "learning_rate": 6.7923728851461955e-06,
+      "loss": 6.0613,
+      "step": 1330
+    },
+    {
+      "epoch": 0.4049255873687391,
+      "grad_norm": 21.823205947875977,
+      "learning_rate": 6.74659123146805e-06,
+      "loss": 3.6511,
+      "step": 1340
+    },
+    {
+      "epoch": 0.4079474201102969,
+      "grad_norm": 15.464320182800293,
+      "learning_rate": 6.70064218061742e-06,
+      "loss": 2.8003,
+      "step": 1350
+    },
+    {
+      "epoch": 0.41096925285185465,
+      "grad_norm": 16.415462493896484,
+      "learning_rate": 6.654530136453119e-06,
+      "loss": 4.3309,
+      "step": 1360
+    },
+    {
+      "epoch": 0.4139910855934124,
+      "grad_norm": 19.546598434448242,
+      "learning_rate": 6.608259518455599e-06,
+      "loss": 5.2213,
+      "step": 1370
+    },
+    {
+      "epoch": 0.4170129183349702,
+      "grad_norm": 14.3574800491333,
+      "learning_rate": 6.5618347613033875e-06,
+      "loss": 5.0989,
+      "step": 1380
+    },
+    {
+      "epoch": 0.42003475107652793,
+      "grad_norm": 16.57396125793457,
+      "learning_rate": 6.5152603144480406e-06,
+      "loss": 5.9716,
+      "step": 1390
+    },
+    {
+      "epoch": 0.4230565838180857,
+      "grad_norm": 30.545312881469727,
+      "learning_rate": 6.468540641687716e-06,
+      "loss": 4.5598,
+      "step": 1400
+    },
+    {
+      "epoch": 0.4260784165596434,
+      "grad_norm": 18.192867279052734,
+      "learning_rate": 6.421680220739337e-06,
+      "loss": 3.9198,
+      "step": 1410
+    },
+    {
+      "epoch": 0.42910024930120116,
+      "grad_norm": 32.15526580810547,
+      "learning_rate": 6.374683542809447e-06,
+      "loss": 7.8433,
+      "step": 1420
+    },
+    {
+      "epoch": 0.4321220820427589,
+      "grad_norm": 15.70085620880127,
+      "learning_rate": 6.327555112163761e-06,
+      "loss": 4.3544,
+      "step": 1430
+    },
+    {
+      "epoch": 0.4351439147843167,
+      "grad_norm": 16.828855514526367,
+      "learning_rate": 6.280299445695469e-06,
+      "loss": 5.1953,
+      "step": 1440
+    },
+    {
+      "epoch": 0.43816574752587445,
+      "grad_norm": 19.912675857543945,
+      "learning_rate": 6.232921072492319e-06,
+      "loss": 4.3359,
+      "step": 1450
+    },
+    {
+      "epoch": 0.4411875802674322,
+      "grad_norm": 25.961816787719727,
+      "learning_rate": 6.185424533402543e-06,
+      "loss": 4.3063,
+      "step": 1460
+    },
+    {
+      "epoch": 0.44420941300899,
+      "grad_norm": 19.82756233215332,
+      "learning_rate": 6.13781438059966e-06,
+      "loss": 3.5112,
+      "step": 1470
+    },
+    {
+      "epoch": 0.44723124575054773,
+      "grad_norm": 33.44493865966797,
+      "learning_rate": 6.090095177146178e-06,
+      "loss": 5.1505,
+      "step": 1480
+    },
+    {
+      "epoch": 0.45025307849210544,
+      "grad_norm": 16.924360275268555,
+      "learning_rate": 6.042271496556255e-06,
+      "loss": 2.8263,
+      "step": 1490
+    },
+    {
+      "epoch": 0.4532749112336632,
+      "grad_norm": 22.583179473876953,
+      "learning_rate": 5.994347922357372e-06,
+      "loss": 3.8395,
+      "step": 1500
+    },
+    {
+      "epoch": 0.45629674397522096,
+      "grad_norm": 18.91628646850586,
+      "learning_rate": 5.946329047651037e-06,
+      "loss": 3.6049,
+      "step": 1510
+    },
+    {
+      "epoch": 0.4593185767167787,
+      "grad_norm": 18.775754928588867,
+      "learning_rate": 5.8982194746725686e-06,
+      "loss": 2.7111,
+      "step": 1520
+    },
+    {
+      "epoch": 0.4623404094583365,
+      "grad_norm": 23.321266174316406,
+      "learning_rate": 5.850023814350007e-06,
+      "loss": 4.2809,
+      "step": 1530
+    },
+    {
+      "epoch": 0.46536224219989425,
+      "grad_norm": 11.822239875793457,
+      "learning_rate": 5.801746685862197e-06,
+      "loss": 6.0749,
+      "step": 1540
+    },
+    {
+      "epoch": 0.468384074941452,
+      "grad_norm": 15.131773948669434,
+      "learning_rate": 5.753392716196069e-06,
+      "loss": 2.8267,
+      "step": 1550
+    },
+    {
+      "epoch": 0.4714059076830098,
+      "grad_norm": 14.163749694824219,
+      "learning_rate": 5.704966539703185e-06,
+      "loss": 3.6323,
+      "step": 1560
+    },
+    {
+      "epoch": 0.4744277404245675,
+      "grad_norm": 16.469493865966797,
+      "learning_rate": 5.656472797655571e-06,
+      "loss": 4.4244,
+      "step": 1570
+    },
+    {
+      "epoch": 0.47744957316612524,
+      "grad_norm": 16.80313491821289,
+      "learning_rate": 5.60791613780088e-06,
+      "loss": 2.7757,
+      "step": 1580
+    },
+    {
+      "epoch": 0.480471405907683,
+      "grad_norm": 22.86004638671875,
+      "learning_rate": 5.5593012139169525e-06,
+      "loss": 4.3241,
+      "step": 1590
+    },
+    {
+      "epoch": 0.48349323864924076,
+      "grad_norm": 21.40558624267578,
+      "learning_rate": 5.510632685365777e-06,
+      "loss": 4.4159,
+      "step": 1600
+    },
+    {
+      "epoch": 0.4865150713907985,
+      "grad_norm": 17.967065811157227,
+      "learning_rate": 5.461915216646938e-06,
+      "loss": 2.8152,
+      "step": 1610
+    },
+    {
+      "epoch": 0.4895369041323563,
+      "grad_norm": 21.66280174255371,
+      "learning_rate": 5.41315347695055e-06,
+      "loss": 4.2485,
+      "step": 1620
+    },
+    {
+      "epoch": 0.49255873687391405,
+      "grad_norm": 21.95995330810547,
+      "learning_rate": 5.364352139709758e-06,
+      "loss": 4.9245,
+      "step": 1630
+    },
+    {
+      "epoch": 0.4955805696154718,
+      "grad_norm": 16.06268310546875,
+      "learning_rate": 5.315515882152822e-06,
+      "loss": 3.5401,
+      "step": 1640
+    },
+    {
+      "epoch": 0.4986024023570295,
+      "grad_norm": 13.898233413696289,
+      "learning_rate": 5.266649384854842e-06,
+      "loss": 4.2544,
+      "step": 1650
+    },
+    {
+      "epoch": 0.5016242350985873,
+      "grad_norm": 21.29293441772461,
+      "learning_rate": 5.217757331289165e-06,
+      "loss": 3.7029,
+      "step": 1660
+    },
+    {
+      "epoch": 0.5046460678401451,
+      "grad_norm": 19.487884521484375,
+      "learning_rate": 5.168844407378506e-06,
+      "loss": 4.8677,
+      "step": 1670
+    },
+    {
+      "epoch": 0.5076679005817029,
+      "grad_norm": 13.403663635253906,
+      "learning_rate": 5.119915301045836e-06,
+      "loss": 2.8783,
+      "step": 1680
+    },
+    {
+      "epoch": 0.5106897333232605,
+      "grad_norm": 23.766260147094727,
+      "learning_rate": 5.070974701765089e-06,
+      "loss": 5.2051,
+      "step": 1690
+    },
+    {
+      "epoch": 0.5137115660648183,
+      "grad_norm": 20.28452491760254,
+      "learning_rate": 5.022027300111712e-06,
+      "loss": 4.4194,
+      "step": 1700
+    },
+    {
+      "epoch": 0.516733398806376,
+      "grad_norm": 15.334933280944824,
+      "learning_rate": 4.973077787313099e-06,
+      "loss": 4.4744,
+      "step": 1710
+    },
+    {
+      "epoch": 0.5197552315479338,
+      "grad_norm": 16.702787399291992,
+      "learning_rate": 4.924130854798983e-06,
+      "loss": 5.1432,
+      "step": 1720
+    },
+    {
+      "epoch": 0.5227770642894916,
+      "grad_norm": 14.432331085205078,
+      "learning_rate": 4.875191193751803e-06,
+      "loss": 2.7788,
+      "step": 1730
+    },
+    {
+      "epoch": 0.5257988970310493,
+      "grad_norm": 15.956453323364258,
+      "learning_rate": 4.826263494657077e-06,
+      "loss": 3.4763,
+      "step": 1740
+    },
+    {
+      "epoch": 0.5288207297726071,
+      "grad_norm": 34.29404830932617,
+      "learning_rate": 4.777352446853863e-06,
+      "loss": 5.0345,
+      "step": 1750
+    },
+    {
+      "epoch": 0.5318425625141648,
+      "grad_norm": 20.15493392944336,
+      "learning_rate": 4.72846273808533e-06,
+      "loss": 3.5034,
+      "step": 1760
+    },
+    {
+      "epoch": 0.5348643952557226,
+      "grad_norm": 18.872026443481445,
+      "learning_rate": 4.679599054049458e-06,
+      "loss": 3.3785,
+      "step": 1770
+    },
+    {
+      "epoch": 0.5378862279972804,
+      "grad_norm": 19.723176956176758,
+      "learning_rate": 4.630766077949965e-06,
+      "loss": 6.0113,
+      "step": 1780
+    },
+    {
+      "epoch": 0.5409080607388381,
+      "grad_norm": 31.925785064697266,
+      "learning_rate": 4.5819684900474484e-06,
+      "loss": 4.3306,
+      "step": 1790
+    },
+    {
+      "epoch": 0.5439298934803959,
+      "grad_norm": 14.491710662841797,
+      "learning_rate": 4.5332109672108245e-06,
+      "loss": 4.4546,
+      "step": 1800
+    },
+    {
+      "epoch": 0.5469517262219536,
+      "grad_norm": 13.951831817626953,
+      "learning_rate": 4.484498182469085e-06,
+      "loss": 3.6289,
+      "step": 1810
+    },
+    {
+      "epoch": 0.5499735589635114,
+      "grad_norm": 20.55958366394043,
+      "learning_rate": 4.435834804563422e-06,
+      "loss": 5.909,
+      "step": 1820
+    },
+    {
+      "epoch": 0.5529953917050692,
+      "grad_norm": 20.23394775390625,
+      "learning_rate": 4.387225497499767e-06,
+      "loss": 3.5712,
+      "step": 1830
+    },
+    {
+      "epoch": 0.5560172244466269,
+      "grad_norm": 16.559497833251953,
+      "learning_rate": 4.3386749201017856e-06,
+      "loss": 3.4663,
+      "step": 1840
+    },
+    {
+      "epoch": 0.5590390571881846,
+      "grad_norm": 15.608750343322754,
+      "learning_rate": 4.290187725564356e-06,
+      "loss": 6.0263,
+      "step": 1850
+    },
+    {
+      "epoch": 0.5620608899297423,
+      "grad_norm": 14.97095775604248,
+      "learning_rate": 4.2417685610076135e-06,
+      "loss": 3.4707,
+      "step": 1860
+    },
+    {
+      "epoch": 0.5650827226713001,
+      "grad_norm": 14.489383697509766,
+      "learning_rate": 4.193422067031535e-06,
+      "loss": 4.3081,
+      "step": 1870
+    },
+    {
+      "epoch": 0.5681045554128579,
+      "grad_norm": 14.916642189025879,
+      "learning_rate": 4.145152877271196e-06,
+      "loss": 4.1373,
+      "step": 1880
+    },
+    {
+      "epoch": 0.5711263881544156,
+      "grad_norm": 16.6586856842041,
+      "learning_rate": 4.096965617952667e-06,
+      "loss": 4.4436,
+      "step": 1890
+    },
+    {
+      "epoch": 0.5741482208959734,
+      "grad_norm": 19.84169578552246,
+      "learning_rate": 4.048864907449619e-06,
+      "loss": 3.4927,
+      "step": 1900
+    },
+    {
+      "epoch": 0.5771700536375312,
+      "grad_norm": 19.735633850097656,
+      "learning_rate": 4.000855355840695e-06,
+      "loss": 3.5524,
+      "step": 1910
+    },
+    {
+      "epoch": 0.5801918863790889,
+      "grad_norm": 17.886032104492188,
+      "learning_rate": 3.952941564467665e-06,
+      "loss": 4.2525,
+      "step": 1920
+    },
+    {
+      "epoch": 0.5832137191206467,
+      "grad_norm": 19.78135871887207,
+      "learning_rate": 3.905128125494427e-06,
+      "loss": 4.3003,
+      "step": 1930
+    },
+    {
+      "epoch": 0.5862355518622044,
+      "grad_norm": 32.764678955078125,
+      "learning_rate": 3.8574196214668876e-06,
+      "loss": 4.5226,
+      "step": 1940
+    },
+    {
+      "epoch": 0.5892573846037622,
+      "grad_norm": 29.241281509399414,
+      "learning_rate": 3.8098206248737486e-06,
+      "loss": 5.1581,
+      "step": 1950
+    },
+    {
+      "epoch": 0.59227921734532,
+      "grad_norm": 21.178043365478516,
+      "learning_rate": 3.7623356977082794e-06,
+      "loss": 2.6207,
+      "step": 1960
+    },
+    {
+      "epoch": 0.5953010500868777,
+      "grad_norm": 21.119003295898438,
+      "learning_rate": 3.714969391031084e-06,
+      "loss": 4.2953,
+      "step": 1970
+    },
+    {
+      "epoch": 0.5983228828284355,
+      "grad_norm": 14.517345428466797,
+      "learning_rate": 3.6677262445339136e-06,
+      "loss": 3.5911,
+      "step": 1980
+    },
+    {
+      "epoch": 0.6013447155699932,
+      "grad_norm": 17.18401527404785,
+      "learning_rate": 3.6206107861045803e-06,
+      "loss": 2.6191,
+      "step": 1990
+    },
+    {
+      "epoch": 0.604366548311551,
+      "grad_norm": 13.34080982208252,
+      "learning_rate": 3.5736275313929826e-06,
+      "loss": 4.3353,
+      "step": 2000
+    },
+    {
+      "epoch": 0.6073883810531087,
+      "grad_norm": 16.087505340576172,
+      "learning_rate": 3.5267809833783213e-06,
+      "loss": 4.8641,
+      "step": 2010
+    },
+    {
+      "epoch": 0.6104102137946664,
+      "grad_norm": 23.290111541748047,
+      "learning_rate": 3.4800756319375326e-06,
+      "loss": 3.52,
+      "step": 2020
+    },
+    {
+      "epoch": 0.6134320465362242,
+      "grad_norm": 22.9693660736084,
+      "learning_rate": 3.433515953414953e-06,
+      "loss": 2.8082,
+      "step": 2030
+    },
+    {
+      "epoch": 0.616453879277782,
+      "grad_norm": 32.13260269165039,
+      "learning_rate": 3.387106410193308e-06,
+      "loss": 5.8238,
+      "step": 2040
+    },
+    {
+      "epoch": 0.6194757120193397,
+      "grad_norm": 20.72865867614746,
+      "learning_rate": 3.3408514502660195e-06,
+      "loss": 5.2489,
+      "step": 2050
+    },
+    {
+      "epoch": 0.6224975447608975,
+      "grad_norm": 15.360533714294434,
+      "learning_rate": 3.2947555068109057e-06,
+      "loss": 3.4271,
+      "step": 2060
+    },
+    {
+      "epoch": 0.6255193775024552,
+      "grad_norm": 18.229646682739258,
+      "learning_rate": 3.248822997765295e-06,
+      "loss": 2.8352,
+      "step": 2070
+    },
+    {
+      "epoch": 0.628541210244013,
+      "grad_norm": 19.263635635375977,
+      "learning_rate": 3.203058325402599e-06,
+      "loss": 4.3576,
+      "step": 2080
+    },
+    {
+      "epoch": 0.6315630429855708,
+      "grad_norm": 18.10948371887207,
+      "learning_rate": 3.1574658759103904e-06,
+      "loss": 4.323,
+      "step": 2090
+    },
+    {
+      "epoch": 0.6345848757271285,
+      "grad_norm": 24.11922264099121,
+      "learning_rate": 3.1120500189700204e-06,
+      "loss": 3.4623,
+      "step": 2100
+    },
+    {
+      "epoch": 0.6376067084686863,
+      "grad_norm": 17.615623474121094,
+      "learning_rate": 3.066815107337815e-06,
+      "loss": 4.239,
+      "step": 2110
+    },
+    {
+      "epoch": 0.640628541210244,
+      "grad_norm": 17.02035140991211,
+      "learning_rate": 3.0217654764279114e-06,
+      "loss": 3.6575,
+      "step": 2120
+    },
+    {
+      "epoch": 0.6436503739518018,
+      "grad_norm": 24.722911834716797,
+      "learning_rate": 2.9769054438967192e-06,
+      "loss": 6.0434,
+      "step": 2130
+    },
+    {
+      "epoch": 0.6466722066933596,
+      "grad_norm": 30.0717830657959,
+      "learning_rate": 2.9322393092291256e-06,
+      "loss": 5.6673,
+      "step": 2140
+    },
+    {
+      "epoch": 0.6496940394349173,
+      "grad_norm": 18.82298469543457,
+      "learning_rate": 2.887771353326422e-06,
+      "loss": 4.2853,
+      "step": 2150
+    },
+    {
+      "epoch": 0.6527158721764751,
+      "grad_norm": 15.994894027709961,
+      "learning_rate": 2.8435058380959957e-06,
+      "loss": 2.6522,
+      "step": 2160
+    },
+    {
+      "epoch": 0.6557377049180327,
+      "grad_norm": 17.15923500061035,
+      "learning_rate": 2.7994470060428835e-06,
+      "loss": 4.1311,
+      "step": 2170
+    },
+    {
+      "epoch": 0.6587595376595905,
+      "grad_norm": 16.892683029174805,
+      "learning_rate": 2.7555990798631436e-06,
+      "loss": 4.9307,
+      "step": 2180
+    },
+    {
+      "epoch": 0.6617813704011483,
+      "grad_norm": 16.922779083251953,
+      "learning_rate": 2.711966262039145e-06,
+      "loss": 3.3275,
+      "step": 2190
+    },
+    {
+      "epoch": 0.664803203142706,
+      "grad_norm": 20.438188552856445,
+      "learning_rate": 2.668552734436802e-06,
+      "loss": 4.3642,
+      "step": 2200
+    },
+    {
+      "epoch": 0.6678250358842638,
+      "grad_norm": 17.198863983154297,
+      "learning_rate": 2.6253626579047653e-06,
+      "loss": 5.8144,
+      "step": 2210
+    },
+    {
+      "epoch": 0.6708468686258215,
+      "grad_norm": 32.332481384277344,
+      "learning_rate": 2.582400171875638e-06,
+      "loss": 3.4467,
+      "step": 2220
+    },
+    {
+      "epoch": 0.6738687013673793,
+      "grad_norm": 28.988691329956055,
+      "learning_rate": 2.5396693939692474e-06,
+      "loss": 4.905,
+      "step": 2230
+    },
+    {
+      "epoch": 0.6768905341089371,
+      "grad_norm": 13.460684776306152,
+      "learning_rate": 2.4971744195979985e-06,
+      "loss": 5.1182,
+      "step": 2240
+    },
+    {
+      "epoch": 0.6799123668504948,
+      "grad_norm": 30.923328399658203,
+      "learning_rate": 2.4549193215743706e-06,
+      "loss": 5.8667,
+      "step": 2250
+    },
+    {
+      "epoch": 0.6829341995920526,
+      "grad_norm": 18.018512725830078,
+      "learning_rate": 2.4129081497205536e-06,
+      "loss": 3.3709,
+      "step": 2260
+    },
+    {
+      "epoch": 0.6859560323336104,
+      "grad_norm": 30.849058151245117,
+      "learning_rate": 2.3711449304803174e-06,
+      "loss": 4.1145,
+      "step": 2270
+    },
+    {
+      "epoch": 0.6889778650751681,
+      "grad_norm": 25.983388900756836,
+      "learning_rate": 2.329633666533103e-06,
+      "loss": 4.0748,
+      "step": 2280
+    },
+    {
+      "epoch": 0.6919996978167259,
+      "grad_norm": 17.239654541015625,
+      "learning_rate": 2.288378336410398e-06,
+      "loss": 4.238,
+      "step": 2290
+    },
+    {
+      "epoch": 0.6950215305582836,
+      "grad_norm": 17.989116668701172,
+      "learning_rate": 2.2473828941144277e-06,
+      "loss": 4.8996,
+      "step": 2300
+    },
+    {
+      "epoch": 0.6980433632998414,
+      "grad_norm": 32.616302490234375,
+      "learning_rate": 2.20665126873919e-06,
+      "loss": 3.3575,
+      "step": 2310
+    },
+    {
+      "epoch": 0.7010651960413992,
+      "grad_norm": 15.997688293457031,
+      "learning_rate": 2.1661873640938818e-06,
+      "loss": 4.1233,
+      "step": 2320
+    },
+    {
+      "epoch": 0.7040870287829568,
+      "grad_norm": 19.421972274780273,
+      "learning_rate": 2.1259950583287633e-06,
+      "loss": 4.1367,
+      "step": 2330
+    },
+    {
+      "epoch": 0.7071088615245146,
+      "grad_norm": 20.1212100982666,
+      "learning_rate": 2.086078203563439e-06,
+      "loss": 4.7622,
+      "step": 2340
+    },
+    {
+      "epoch": 0.7101306942660723,
+      "grad_norm": 14.532552719116211,
+      "learning_rate": 2.0464406255176967e-06,
+      "loss": 4.0196,
+      "step": 2350
+    },
+    {
+      "epoch": 0.7131525270076301,
+      "grad_norm": 29.55924415588379,
+      "learning_rate": 2.0070861231448142e-06,
+      "loss": 4.9694,
+      "step": 2360
+    },
+    {
+      "epoch": 0.7161743597491879,
+      "grad_norm": 21.097187042236328,
+      "learning_rate": 1.968018468267472e-06,
+      "loss": 4.2314,
+      "step": 2370
+    },
+    {
+      "epoch": 0.7191961924907456,
+      "grad_norm": 13.943464279174805,
+      "learning_rate": 1.929241405216254e-06,
+      "loss": 4.0535,
+      "step": 2380
+    },
+    {
+      "epoch": 0.7222180252323034,
+      "grad_norm": 19.29151725769043,
+      "learning_rate": 1.8907586504707776e-06,
+      "loss": 4.7497,
+      "step": 2390
+    },
+    {
+      "epoch": 0.7252398579738611,
+      "grad_norm": 18.542991638183594,
+      "learning_rate": 1.8525738923035002e-06,
+      "loss": 2.5704,
+      "step": 2400
+    },
+    {
+      "epoch": 0.7282616907154189,
+      "grad_norm": 18.23619842529297,
+      "learning_rate": 1.8146907904262268e-06,
+      "loss": 4.2542,
+      "step": 2410
+    },
+    {
+      "epoch": 0.7312835234569767,
+      "grad_norm": 21.722688674926758,
+      "learning_rate": 1.7771129756393545e-06,
+      "loss": 3.4271,
+      "step": 2420
+    },
+    {
+      "epoch": 0.7343053561985344,
+      "grad_norm": 16.661977767944336,
+      "learning_rate": 1.7398440494838947e-06,
+      "loss": 3.604,
+      "step": 2430
+    },
+    {
+      "epoch": 0.7373271889400922,
+      "grad_norm": 16.67827606201172,
+      "learning_rate": 1.7028875838962822e-06,
+      "loss": 4.1239,
+      "step": 2440
+    },
+    {
+      "epoch": 0.74034902168165,
+      "grad_norm": 33.99874496459961,
+      "learning_rate": 1.6662471208660392e-06,
+      "loss": 4.0579,
+      "step": 2450
+    },
+    {
+      "epoch": 0.7433708544232077,
+      "grad_norm": 21.555774688720703,
+      "learning_rate": 1.6299261720963095e-06,
+      "loss": 4.1917,
+      "step": 2460
+    },
+    {
+      "epoch": 0.7463926871647655,
+      "grad_norm": 14.352119445800781,
+      "learning_rate": 1.5939282186672705e-06,
+      "loss": 4.9304,
+      "step": 2470
+    },
+    {
+      "epoch": 0.7494145199063232,
+      "grad_norm": 20.739620208740234,
+      "learning_rate": 1.5582567107025237e-06,
+      "loss": 4.8278,
+      "step": 2480
+    },
+    {
+      "epoch": 0.7524363526478809,
+      "grad_norm": 29.204471588134766,
+      "learning_rate": 1.5229150670384057e-06,
+      "loss": 3.3692,
+      "step": 2490
+    },
+    {
+      "epoch": 0.7554581853894387,
+      "grad_norm": 20.03524398803711,
+      "learning_rate": 1.4879066748963295e-06,
+      "loss": 2.5494,
+      "step": 2500
+    },
+    {
+      "epoch": 0.7584800181309964,
+      "grad_norm": 27.458927154541016,
+      "learning_rate": 1.4532348895581466e-06,
+      "loss": 3.4744,
+      "step": 2510
+    },
+    {
+      "epoch": 0.7615018508725542,
+      "grad_norm": 28.542675018310547,
+      "learning_rate": 1.4189030340445648e-06,
+      "loss": 6.7224,
+      "step": 2520
+    },
+    {
+      "epoch": 0.7645236836141119,
+      "grad_norm": 15.871747016906738,
+      "learning_rate": 1.3849143987966646e-06,
+      "loss": 4.9879,
+      "step": 2530
+    },
+    {
+      "epoch": 0.7675455163556697,
+      "grad_norm": 16.10350799560547,
+      "learning_rate": 1.3512722413605356e-06,
+      "loss": 4.0848,
+      "step": 2540
+    },
+    {
+      "epoch": 0.7705673490972275,
+      "grad_norm": 15.013642311096191,
+      "learning_rate": 1.3179797860750654e-06,
+      "loss": 4.9059,
+      "step": 2550
+    },
+    {
+      "epoch": 0.7735891818387852,
+      "grad_norm": 17.056211471557617,
+      "learning_rate": 1.2850402237629184e-06,
+      "loss": 4.1558,
+      "step": 2560
+    },
+    {
+      "epoch": 0.776611014580343,
+      "grad_norm": 13.357272148132324,
+      "learning_rate": 1.2524567114247083e-06,
+      "loss": 3.3524,
+      "step": 2570
+    },
+    {
+      "epoch": 0.7796328473219007,
+      "grad_norm": 15.318449974060059,
+      "learning_rate": 1.2202323719364324e-06,
+      "loss": 3.2666,
+      "step": 2580
+    },
+    {
+      "epoch": 0.7826546800634585,
+      "grad_norm": 20.867525100708008,
+      "learning_rate": 1.1883702937501708e-06,
+      "loss": 4.1267,
+      "step": 2590
+    },
+    {
+      "epoch": 0.7856765128050163,
+      "grad_norm": 20.91588020324707,
+      "learning_rate": 1.1568735305980694e-06,
+      "loss": 4.1263,
+      "step": 2600
+    },
+    {
+      "epoch": 0.788698345546574,
+      "grad_norm": 23.296667098999023,
+      "learning_rate": 1.1257451011996807e-06,
+      "loss": 3.4198,
+      "step": 2610
+    },
+    {
+      "epoch": 0.7917201782881318,
+      "grad_norm": 32.17024230957031,
+      "learning_rate": 1.0949879889726295e-06,
+      "loss": 3.4727,
+      "step": 2620
+    },
+    {
+      "epoch": 0.7947420110296896,
+      "grad_norm": 30.94739532470703,
+      "learning_rate": 1.0646051417466801e-06,
+      "loss": 3.4438,
+      "step": 2630
+    },
+    {
+      "epoch": 0.7977638437712473,
+      "grad_norm": 24.171409606933594,
+      "learning_rate": 1.0345994714812135e-06,
+      "loss": 4.1562,
+      "step": 2640
+    },
+    {
+      "epoch": 0.800785676512805,
+      "grad_norm": 15.973287582397461,
+      "learning_rate": 1.0049738539861332e-06,
+      "loss": 3.3106,
+      "step": 2650
+    },
+    {
+      "epoch": 0.8038075092543627,
+      "grad_norm": 16.17388343811035,
+      "learning_rate": 9.757311286462428e-07,
+      "loss": 4.1695,
+      "step": 2660
+    },
+    {
+      "epoch": 0.8068293419959205,
+      "grad_norm": 19.307254791259766,
+      "learning_rate": 9.468740981491143e-07,
+      "loss": 4.2039,
+      "step": 2670
+    },
+    {
+      "epoch": 0.8098511747374783,
+      "grad_norm": 13.707403182983398,
+      "learning_rate": 9.1840552821647e-07,
+      "loss": 4.0537,
+      "step": 2680
+    },
+    {
+      "epoch": 0.812873007479036,
+      "grad_norm": 16.229114532470703,
+      "learning_rate": 8.903281473391152e-07,
+      "loss": 3.3216,
+      "step": 2690
+    },
+    {
+      "epoch": 0.8158948402205938,
+      "grad_norm": 16.56165313720703,
+      "learning_rate": 8.62644646515427e-07,
+      "loss": 5.7495,
+      "step": 2700
+    },
+    {
+      "epoch": 0.8189166729621515,
+      "grad_norm": 13.074578285217285,
+      "learning_rate": 8.353576789934436e-07,
+      "loss": 3.4104,
+      "step": 2710
+    },
+    {
+      "epoch": 0.8219385057037093,
+      "grad_norm": 20.466753005981445,
+      "learning_rate": 8.084698600165797e-07,
+      "loss": 3.5412,
+      "step": 2720
+    },
+    {
+      "epoch": 0.8249603384452671,
+      "grad_norm": 18.137537002563477,
+      "learning_rate": 7.819837665729596e-07,
+      "loss": 4.105,
+      "step": 2730
+    },
+    {
+      "epoch": 0.8279821711868248,
+      "grad_norm": 28.85208511352539,
+      "learning_rate": 7.559019371484521e-07,
+      "loss": 3.3502,
+      "step": 2740
+    },
+    {
+      "epoch": 0.8310040039283826,
+      "grad_norm": 16.86795997619629,
+      "learning_rate": 7.302268714833622e-07,
+      "loss": 4.1471,
+      "step": 2750
+    },
+    {
+      "epoch": 0.8340258366699403,
+      "grad_norm": 16.45426368713379,
+      "learning_rate": 7.049610303328541e-07,
+      "loss": 3.531,
+      "step": 2760
+    },
+    {
+      "epoch": 0.8370476694114981,
+      "grad_norm": 16.50694465637207,
+      "learning_rate": 6.80106835231113e-07,
+      "loss": 4.2983,
+      "step": 2770
+    },
+    {
+      "epoch": 0.8400695021530559,
+      "grad_norm": 18.195581436157227,
+      "learning_rate": 6.556666682592494e-07,
+      "loss": 3.3125,
+      "step": 2780
+    },
+    {
+      "epoch": 0.8430913348946136,
+      "grad_norm": 16.14658546447754,
+      "learning_rate": 6.316428718170037e-07,
+      "loss": 2.4715,
+      "step": 2790
+    },
+    {
+      "epoch": 0.8461131676361714,
+      "grad_norm": 19.641435623168945,
+      "learning_rate": 6.080377483982425e-07,
+      "loss": 3.2894,
+      "step": 2800
+    },
+    {
+      "epoch": 0.849135000377729,
+      "grad_norm": 18.186119079589844,
+      "learning_rate": 5.848535603702798e-07,
+      "loss": 3.3862,
+      "step": 2810
+    },
+    {
+      "epoch": 0.8521568331192868,
+      "grad_norm": 15.75075912475586,
+      "learning_rate": 5.62092529757054e-07,
+      "loss": 6.3941,
+      "step": 2820
+    },
+    {
+      "epoch": 0.8551786658608446,
+      "grad_norm": 20.051259994506836,
+      "learning_rate": 5.397568380261559e-07,
+      "loss": 2.3455,
+      "step": 2830
+    },
+    {
+      "epoch": 0.8582004986024023,
+      "grad_norm": 15.361984252929688,
+      "learning_rate": 5.178486258797555e-07,
+      "loss": 4.1372,
+      "step": 2840
+    },
+    {
+      "epoch": 0.8612223313439601,
+      "grad_norm": 15.885826110839844,
+      "learning_rate": 4.963699930494365e-07,
+      "loss": 3.3724,
+      "step": 2850
+    },
+    {
+      "epoch": 0.8642441640855179,
+      "grad_norm": 20.120771408081055,
+      "learning_rate": 4.75322998094942e-07,
+      "loss": 4.2862,
+      "step": 2860
+    },
+    {
+      "epoch": 0.8672659968270756,
+      "grad_norm": 19.651308059692383,
+      "learning_rate": 4.5470965820689384e-07,
+      "loss": 2.6081,
+      "step": 2870
+    },
+    {
+      "epoch": 0.8702878295686334,
+      "grad_norm": 20.324071884155273,
+      "learning_rate": 4.345319490134453e-07,
+      "loss": 3.0586,
+      "step": 2880
+    },
+    {
+      "epoch": 0.8733096623101911,
+      "grad_norm": 15.238956451416016,
+      "learning_rate": 4.147918043909405e-07,
+      "loss": 4.7833,
+      "step": 2890
+    },
+    {
+      "epoch": 0.8763314950517489,
+      "grad_norm": 17.426593780517578,
+      "learning_rate": 3.9549111627856794e-07,
+      "loss": 4.7313,
+      "step": 2900
+    },
+    {
+      "epoch": 0.8793533277933067,
+      "grad_norm": 17.20795440673828,
+      "learning_rate": 3.766317344970288e-07,
+      "loss": 4.1551,
+      "step": 2910
+    },
+    {
+      "epoch": 0.8823751605348644,
+      "grad_norm": 18.264366149902344,
+      "learning_rate": 3.582154665712473e-07,
+      "loss": 2.485,
+      "step": 2920
+    },
+    {
+      "epoch": 0.8853969932764222,
+      "grad_norm": 20.325515747070312,
+      "learning_rate": 3.402440775571364e-07,
+      "loss": 4.06,
+      "step": 2930
+    },
+    {
+      "epoch": 0.88841882601798,
+      "grad_norm": 24.280147552490234,
+      "learning_rate": 3.227192898724252e-07,
+      "loss": 5.7449,
+      "step": 2940
+    },
+    {
+      "epoch": 0.8914406587595377,
+      "grad_norm": 16.266826629638672,
+      "learning_rate": 3.056427831315878e-07,
+      "loss": 3.3582,
+      "step": 2950
+    },
+    {
+      "epoch": 0.8944624915010955,
+      "grad_norm": 24.630563735961914,
+      "learning_rate": 2.890161939848535e-07,
+      "loss": 4.1681,
+      "step": 2960
+    },
+    {
+      "epoch": 0.8974843242426531,
+      "grad_norm": 17.58610725402832,
+      "learning_rate": 2.72841115961357e-07,
+      "loss": 4.2704,
+      "step": 2970
+    },
+    {
+      "epoch": 0.9005061569842109,
+      "grad_norm": 19.43164825439453,
+      "learning_rate": 2.5711909931640633e-07,
+      "loss": 4.0315,
+      "step": 2980
+    },
+    {
+      "epoch": 0.9035279897257686,
+      "grad_norm": 20.064899444580078,
+      "learning_rate": 2.418516508829e-07,
+      "loss": 2.5202,
+      "step": 2990
+    },
+    {
+      "epoch": 0.9065498224673264,
+      "grad_norm": 31.959165573120117,
+      "learning_rate": 2.270402339269162e-07,
+      "loss": 5.6536,
+      "step": 3000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 3309,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0651141456648995e+19,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoints/E118/checkpoint-3000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d47c326ae209c5cfdf45f2fbefa74578242ebc5f1af8b4ec289f620cf6f14086
+size 7697

checkpoints/E118/checkpoint-3000/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)