Upload folder using huggingface_hub

config.json

@@ -0,0 +1,70 @@
+{
+  "action_horizon": 50,
+  "add_pos_embed": true,
+  "apply_sincos_state_encoding": true,
+  "architectures": [
+    "Gr00tN1d6"
+  ],
+  "attn_dropout": 0.2,
+  "attn_implementation": null,
+  "backbone_embedding_dim": 2048,
+  "backbone_model_type": "eagle",
+  "backbone_trainable_params_fp32": true,
+  "collator_overwrite_image_inputs": false,
+  "color_jitter_params": {
+    "brightness": 0.1,
+    "contrast": 0.1,
+    "hue": 0.1,
+    "saturation": 0.1
+  },
+  "crop_fraction": 0.95,
+  "diffusion_model_cfg": {
+    "attention_head_dim": 48,
+    "dropout": 0.2,
+    "final_dropout": true,
+    "interleave_self_attention": true,
+    "norm_type": "ada_norm",
+    "num_attention_heads": 32,
+    "num_layers": 32,
+    "output_dim": 1024,
+    "positional_embeddings": null
+  },
+  "eagle_collator": true,
+  "formalize_language": true,
+  "gemma_collator": false,
+  "hidden_size": 1024,
+  "image_crop_size": null,
+  "image_target_size": null,
+  "input_embedding_dim": 1536,
+  "load_bf16": true,
+  "max_action_dim": 128,
+  "max_num_embodiments": 32,
+  "max_seq_len": 1024,
+  "max_state_dim": 128,
+  "model_dtype": "bfloat16",
+  "model_name": "nvidia/Eagle-Block2A-2B-v2",
+  "model_type": "Gr00tN1d6",
+  "noise_beta_alpha": 1.5,
+  "noise_beta_beta": 1.0,
+  "noise_s": 0.999,
+  "num_inference_timesteps": 4,
+  "num_timestep_buckets": 1000,
+  "random_rotation_angle": null,
+  "reproject_vision": false,
+  "select_layer": 16,
+  "shortest_image_edge": 256,
+  "state_dropout_prob": 0.0,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "tune_diffusion_model": true,
+  "tune_llm": false,
+  "tune_projector": true,
+  "tune_top_llm_layers": 4,
+  "tune_visual": false,
+  "tune_vlln": true,
+  "use_albumentations_transforms": true,
+  "use_alternate_vl_dit": true,
+  "use_flash_attention": true,
+  "use_relative_action": true,
+  "use_vlln": true
+}

embodiment_id.json

@@ -0,0 +1,11 @@
+{
+  "robocasa_panda_omron": 13,
+  "gr1": 20,
+  "behavior_r1_pro": 24,
+  "unitree_g1": 8,
+  "oxe_google": 0,
+  "oxe_widowx": 1,
+  "libero_panda": 2,
+  "oxe_droid": 16,
+  "new_embodiment": 10
+}

experiment_cfg/conf.yaml

@@ -0,0 +1,253 @@
+load_config_path: null
+model:
+  model_type: Gr00tN1d6
+  model_dtype: bfloat16
+  model_name: nvidia/Eagle-Block2A-2B-v2
+  backbone_model_type: eagle
+  model_revision: null
+  tune_top_llm_layers: 4
+  backbone_embedding_dim: 2048
+  tune_llm: false
+  tune_visual: false
+  select_layer: 16
+  reproject_vision: false
+  use_flash_attention: true
+  load_bf16: false
+  collator_overwrite_image_inputs: false
+  eagle_collator: true
+  backbone_trainable_params_fp32: true
+  image_crop_size: null
+  image_target_size: null
+  shortest_image_edge: 256
+  crop_fraction: 0.95
+  random_rotation_angle: null
+  color_jitter_params:
+    brightness: 0.3
+    contrast: 0.4
+    saturation: 0.5
+    hue: 0.08
+  use_albumentations_transforms: true
+  extra_augmentation_config: null
+  formalize_language: true
+  apply_sincos_state_encoding: false
+  use_relative_action: true
+  max_state_dim: 29
+  max_action_dim: 29
+  action_horizon: 16
+  hidden_size: 1024
+  input_embedding_dim: 1536
+  add_pos_embed: true
+  attn_dropout: 0.2
+  use_vlln: true
+  max_seq_len: 1024
+  use_alternate_vl_dit: true
+  attend_text_every_n_blocks: 2
+  diffusion_model_cfg:
+    positional_embeddings: null
+    num_layers: 32
+    num_attention_heads: 32
+    attention_head_dim: 48
+    norm_type: ada_norm
+    dropout: 0.2
+    final_dropout: true
+    output_dim: 1024
+    interleave_self_attention: true
+  num_inference_timesteps: 4
+  noise_beta_alpha: 1.5
+  noise_beta_beta: 1.0
+  noise_s: 0.999
+  num_timestep_buckets: 1000
+  tune_projector: true
+  tune_diffusion_model: true
+  tune_vlln: true
+  state_dropout_prob: 0.0
+  state_additive_noise_scale: 0.0
+  max_num_embodiments: 32
+data:
+  datasets:
+  - dataset_paths:
+    - ./datasets/push_box_real_11
+    embodiment_tag: unitree_g1
+    mix_ratio: 1.0
+    dataset_type: physical_embodiment
+    val_dataset_path: null
+  modality_configs:
+    unitree_g1:
+      video:
+        delta_indices:
+        - 0
+        modality_keys:
+        - ego_view
+        sin_cos_embedding_keys: null
+        mean_std_embedding_keys: null
+        action_configs: null
+      state:
+        delta_indices:
+        - 0
+        modality_keys:
+        - left_leg
+        - right_leg
+        - waist
+        - left_arm
+        - right_arm
+        - left_hand
+        - right_hand
+        sin_cos_embedding_keys: null
+        mean_std_embedding_keys: null
+        action_configs: null
+      action:
+        delta_indices:
+        - 0
+        - 1
+        - 2
+        - 3
+        - 4
+        - 5
+        - 6
+        - 7
+        - 8
+        - 9
+        - 10
+        - 11
+        - 12
+        - 13
+        - 14
+        - 15
+        - 16
+        - 17
+        - 18
+        - 19
+        - 20
+        - 21
+        - 22
+        - 23
+        - 24
+        - 25
+        - 26
+        - 27
+        - 28
+        - 29
+        modality_keys:
+        - left_arm
+        - right_arm
+        - left_hand
+        - right_hand
+        - waist
+        - base_height_command
+        - navigate_command
+        sin_cos_embedding_keys: null
+        mean_std_embedding_keys: null
+        action_configs:
+        - rep: RELATIVE
+          type: NON_EEF
+          format: DEFAULT
+          state_key: null
+        - rep: RELATIVE
+          type: NON_EEF
+          format: DEFAULT
+          state_key: null
+        - rep: ABSOLUTE
+          type: NON_EEF
+          format: DEFAULT
+          state_key: null
+        - rep: ABSOLUTE
+          type: NON_EEF
+          format: DEFAULT
+          state_key: null
+        - rep: ABSOLUTE
+          type: NON_EEF
+          format: DEFAULT
+          state_key: null
+        - rep: ABSOLUTE
+          type: NON_EEF
+          format: DEFAULT
+          state_key: null
+        - rep: ABSOLUTE
+          type: NON_EEF
+          format: DEFAULT
+          state_key: null
+      language:
+        delta_indices:
+        - 0
+        modality_keys:
+        - annotation.human.task_description
+        sin_cos_embedding_keys: null
+        mean_std_embedding_keys: null
+        action_configs: null
+  download_cache: false
+  shard_size: 1024
+  episode_sampling_rate: 0.1
+  num_shards_per_epoch: 100000
+  override_pretraining_statistics: false
+  mode: single_turn
+  random_chop: 0.0
+  mock_dataset_mode: false
+  shuffle: true
+  seed: 42
+  multiprocessing_context: fork
+  allow_padding: false
+  subsample_ratio: 1.0
+  image_crop_size:
+  - 244
+  - 244
+  image_target_size:
+  - 224
+  - 224
+  video_backend: torchcodec
+training:
+  output_dir: ./outputs
+  experiment_name: null
+  max_steps: 5000
+  global_batch_size: 768
+  batch_size: null
+  gradient_accumulation_steps: 1
+  learning_rate: 0.0001
+  lr_scheduler_type: cosine
+  weight_decay: 1.0e-05
+  warmup_ratio: 0.05
+  warmup_steps: 0
+  max_grad_norm: 1.0
+  optim: adamw_torch
+  start_from_checkpoint: nvidia/GR00T-N1.6-3B
+  tf32: true
+  fp16: false
+  bf16: true
+  eval_bf16: true
+  logging_steps: 10
+  save_steps: 1000
+  save_total_limit: 5
+  save_vl_model: false
+  upload_checkpoints: false
+  upload_every: 1000
+  upload_last_n_checkpoints: 5
+  max_concurrent_uploads: 2
+  eval_strategy: 'no'
+  eval_steps: 500
+  eval_set_split_ratio: 0.1
+  eval_batch_size: 2
+  save_best_eval_metric_name: ''
+  save_best_eval_metric_greater_is_better: true
+  deepspeed_stage: 2
+  gradient_checkpointing: false
+  transformers_trust_remote_code: true
+  transformers_local_files_only: false
+  transformers_cache_dir: null
+  transformers_access_token: null
+  use_ddp: false
+  ddp_bucket_cap_mb: 100
+  num_gpus: 8
+  dataloader_num_workers: 6
+  remove_unused_columns: false
+  use_wandb: true
+  wandb_project: finetune-gr00t-n1d6
+  enable_profiling: false
+  max_retries: 3
+  assert_loss_less_than: null
+  add_rl_callback: false
+  enable_open_loop_eval: false
+  open_loop_eval_traj_ids:
+  - 0
+  open_loop_eval_steps_per_traj: 100
+  open_loop_eval_plot_indices: null
+max_steps: 5000
+save_steps: 1000

experiment_cfg/config.yaml

@@ -0,0 +1,292 @@
+!!python/object:gr00t.configs.base_config.Config
+data: !!python/object:gr00t.configs.data.data_config.DataConfig
+  allow_padding: false
+  datasets:
+  - !!python/object:gr00t.configs.data.data_config.SingleDatasetConfig
+    dataset_paths:
+    - ./datasets/push_box_real_11
+    dataset_type: physical_embodiment
+    embodiment_tag: unitree_g1
+    mix_ratio: 1.0
+    val_dataset_path: null
+  download_cache: false
+  episode_sampling_rate: 0.1
+  image_crop_size:
+  - 244
+  - 244
+  image_target_size:
+  - 224
+  - 224
+  mock_dataset_mode: false
+  modality_configs:
+    unitree_g1:
+      action: !!python/object:gr00t.data.types.ModalityConfig
+        action_configs:
+        - !!python/object:gr00t.data.types.ActionConfig
+          format: &id001 !!python/object/apply:gr00t.data.types.ActionFormat
+          - default
+          rep: &id002 !!python/object/apply:gr00t.data.types.ActionRepresentation
+          - relative
+          state_key: null
+          type: &id003 !!python/object/apply:gr00t.data.types.ActionType
+          - non_eef
+        - !!python/object:gr00t.data.types.ActionConfig
+          format: *id001
+          rep: *id002
+          state_key: null
+          type: *id003
+        - !!python/object:gr00t.data.types.ActionConfig
+          format: *id001
+          rep: &id004 !!python/object/apply:gr00t.data.types.ActionRepresentation
+          - absolute
+          state_key: null
+          type: *id003
+        - !!python/object:gr00t.data.types.ActionConfig
+          format: *id001
+          rep: *id004
+          state_key: null
+          type: *id003
+        - !!python/object:gr00t.data.types.ActionConfig
+          format: *id001
+          rep: *id004
+          state_key: null
+          type: *id003
+        - !!python/object:gr00t.data.types.ActionConfig
+          format: *id001
+          rep: *id004
+          state_key: null
+          type: *id003
+        - !!python/object:gr00t.data.types.ActionConfig
+          format: *id001
+          rep: *id004
+          state_key: null
+          type: *id003
+        delta_indices:
+        - 0
+        - 1
+        - 2
+        - 3
+        - 4
+        - 5
+        - 6
+        - 7
+        - 8
+        - 9
+        - 10
+        - 11
+        - 12
+        - 13
+        - 14
+        - 15
+        - 16
+        - 17
+        - 18
+        - 19
+        - 20
+        - 21
+        - 22
+        - 23
+        - 24
+        - 25
+        - 26
+        - 27
+        - 28
+        - 29
+        mean_std_embedding_keys: null
+        modality_keys:
+        - left_arm
+        - right_arm
+        - left_hand
+        - right_hand
+        - waist
+        - base_height_command
+        - navigate_command
+        sin_cos_embedding_keys: null
+      language: !!python/object:gr00t.data.types.ModalityConfig
+        action_configs: null
+        delta_indices:
+        - 0
+        mean_std_embedding_keys: null
+        modality_keys:
+        - annotation.human.task_description
+        sin_cos_embedding_keys: null
+      state: !!python/object:gr00t.data.types.ModalityConfig
+        action_configs: null
+        delta_indices:
+        - 0
+        mean_std_embedding_keys: null
+        modality_keys:
+        - left_leg
+        - right_leg
+        - waist
+        - left_arm
+        - right_arm
+        - left_hand
+        - right_hand
+        sin_cos_embedding_keys: null
+      video: !!python/object:gr00t.data.types.ModalityConfig
+        action_configs: null
+        delta_indices:
+        - 0
+        mean_std_embedding_keys: null
+        modality_keys:
+        - ego_view
+        sin_cos_embedding_keys: null
+  mode: single_turn
+  multiprocessing_context: fork
+  num_shards_per_epoch: 100000
+  override_pretraining_statistics: false
+  random_chop: 0.0
+  seed: 42
+  shard_size: 1024
+  shuffle: true
+  subsample_ratio: 1.0
+  video_backend: torchcodec
+load_config_path: null
+model: !!python/object:gr00t.configs.model.gr00t_n1d6.Gr00tN1d6Config
+  _attn_implementation_autoset: false
+  _attn_implementation_internal: null
+  _commit_hash: null
+  _name_or_path: ''
+  add_cross_attention: false
+  architectures: null
+  backbone_model_type: eagle
+  backbone_trainable_params_fp32: true
+  bad_words_ids: null
+  begin_suppress_tokens: null
+  bos_token_id: null
+  chunk_size_feed_forward: 0
+  color_jitter_params:
+    brightness: 0.3
+    contrast: 0.4
+    hue: 0.08
+    saturation: 0.5
+  cross_attention_hidden_size: null
+  decoder_start_token_id: null
+  diffusion_model_cfg:
+    attention_head_dim: 48
+    dropout: 0.2
+    final_dropout: true
+    interleave_self_attention: true
+    norm_type: ada_norm
+    num_attention_heads: 32
+    num_layers: 32
+    output_dim: 1024
+    positional_embeddings: null
+  diversity_penalty: 0.0
+  do_sample: false
+  eagle_collator: true
+  early_stopping: false
+  encoder_no_repeat_ngram_size: 0
+  eos_token_id: null
+  exponential_decay_length_penalty: null
+  extra_augmentation_config: null
+  finetuning_task: null
+  forced_bos_token_id: null
+  forced_eos_token_id: null
+  id2label:
+    0: LABEL_0
+    1: LABEL_1
+  is_decoder: false
+  is_encoder_decoder: false
+  label2id:
+    LABEL_0: 0
+    LABEL_1: 1
+  length_penalty: 1.0
+  load_bf16: false
+  max_length: 20
+  min_length: 0
+  model_name: nvidia/Eagle-Block2A-2B-v2
+  no_repeat_ngram_size: 0
+  num_beam_groups: 1
+  num_beams: 1
+  num_return_sequences: 1
+  output_attentions: false
+  output_hidden_states: false
+  output_scores: false
+  pad_token_id: null
+  prefix: null
+  problem_type: null
+  pruned_heads: {}
+  random_rotation_angle: null
+  remove_invalid_values: false
+  repetition_penalty: 1.0
+  reproject_vision: false
+  return_dict: true
+  return_dict_in_generate: false
+  sep_token_id: null
+  state_dropout_prob: 0.0
+  suppress_tokens: null
+  task_specific_params: null
+  temperature: 1.0
+  tf_legacy_loss: false
+  tie_encoder_decoder: false
+  tie_word_embeddings: true
+  tokenizer_class: null
+  top_k: 50
+  top_p: 1.0
+  torch_dtype: null
+  torchscript: false
+  transformers_version: null
+  tune_diffusion_model: true
+  tune_llm: false
+  tune_projector: true
+  tune_visual: false
+  typical_p: 1.0
+  use_bfloat16: false
+  use_relative_action: true
+training: !!python/object:gr00t.configs.training.training_config.TrainingConfig
+  add_rl_callback: false
+  assert_loss_less_than: null
+  batch_size: null
+  bf16: true
+  dataloader_num_workers: 6
+  ddp_bucket_cap_mb: 100
+  deepspeed_stage: 2
+  enable_open_loop_eval: false
+  enable_profiling: false
+  eval_batch_size: 2
+  eval_bf16: true
+  eval_set_split_ratio: 0.1
+  eval_steps: 500
+  eval_strategy: 'no'
+  experiment_name: null
+  fp16: false
+  global_batch_size: 768
+  gradient_accumulation_steps: 1
+  gradient_checkpointing: false
+  learning_rate: 0.0001
+  logging_steps: 10
+  lr_scheduler_type: cosine
+  max_concurrent_uploads: 2
+  max_grad_norm: 1.0
+  max_retries: 3
+  max_steps: 5000
+  num_gpus: 8
+  open_loop_eval_plot_indices: null
+  open_loop_eval_steps_per_traj: 100
+  open_loop_eval_traj_ids:
+  - 0
+  optim: adamw_torch
+  output_dir: ./outputs
+  remove_unused_columns: false
+  save_best_eval_metric_greater_is_better: true
+  save_best_eval_metric_name: ''
+  save_steps: 1000
+  save_total_limit: 5
+  save_vl_model: false
+  start_from_checkpoint: nvidia/GR00T-N1.6-3B
+  tf32: true
+  transformers_access_token: null
+  transformers_cache_dir: null
+  transformers_local_files_only: false
+  transformers_trust_remote_code: true
+  upload_checkpoints: false
+  upload_every: 1000
+  upload_last_n_checkpoints: 5
+  use_ddp: false
+  use_wandb: true
+  wandb_project: finetune-gr00t-n1d6
+  warmup_ratio: 0.05
+  warmup_steps: 0
+  weight_decay: 1.0e-05

experiment_cfg/final_model_config.json

@@ -0,0 +1,54 @@
+{
+  "model_type": "Gr00tN1d6",
+  "model_dtype": "bfloat16",
+  "model_name": "nvidia/Eagle-Block2A-2B-v2",
+  "backbone_model_type": "eagle",
+  "model_revision": null,
+  "tune_top_llm_layers": 4,
+  "backbone_embedding_dim": 2048,
+  "tune_llm": false,
+  "tune_visual": false,
+  "select_layer": 16,
+  "reproject_vision": false,
+  "use_flash_attention": true,
+  "load_bf16": true,
+  "collator_overwrite_image_inputs": false,
+  "eagle_collator": true,
+  "backbone_trainable_params_fp32": true,
+  "extra_augmentation_config": null,
+  "apply_sincos_state_encoding": true,
+  "use_relative_action": true,
+  "max_state_dim": 128,
+  "max_action_dim": 128,
+  "action_horizon": 50,
+  "hidden_size": 1024,
+  "input_embedding_dim": 1536,
+  "add_pos_embed": true,
+  "attn_dropout": 0.2,
+  "use_vlln": true,
+  "max_seq_len": 1024,
+  "use_alternate_vl_dit": true,
+  "attend_text_every_n_blocks": 2,
+  "diffusion_model_cfg": {
+    "attention_head_dim": 48,
+    "dropout": 0.2,
+    "final_dropout": true,
+    "interleave_self_attention": true,
+    "norm_type": "ada_norm",
+    "num_attention_heads": 32,
+    "num_layers": 32,
+    "output_dim": 1024,
+    "positional_embeddings": null
+  },
+  "num_inference_timesteps": 4,
+  "noise_beta_alpha": 1.5,
+  "noise_beta_beta": 1.0,
+  "noise_s": 0.999,
+  "num_timestep_buckets": 1000,
+  "tune_projector": true,
+  "tune_diffusion_model": true,
+  "tune_vlln": true,
+  "state_dropout_prob": 0.0,
+  "state_additive_noise_scale": 0.0,
+  "max_num_embodiments": 32
+}

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 sha256:2bee2ad0ca183018f94471159e9cc04178dc5dba256022b70b79c952329b9ca9
+size 2429964677

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 sha256:3ff0cf1757feafc5bc9cecb4c514dee4896624dc1c368c8879f5d1a0c2b83c6a
+size 2429964421

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 sha256:14403243e919f12ed35e34ad2a102e941ed701b45e3559f44d1684e36f437f9c
+size 2429964613

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:b6815ef723aba74053387a977af3406f590aa4656cd376c570be26bac46a228e
+size 2429964485

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:ff16936c49d3df8ea7dedae3eeec5a112bc07b7772999201f445fe2f06bdee04
+size 2429964421

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 sha256:10388becbee530c724bd170f404bc3eec747b4baea5782b12108fbaf850b6ee9
+size 2429964613

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 sha256:11ac8f8c3c5689ac3f0d50cad84fbf978957b1b78bcc12150256eb27da7810ec
+size 2429962437

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:a8a623a748f99d776d5435a56c20731a14905fb703b1989e3840adcd66e0b14d
+size 2429960773

global_step3000/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5b6f8e52f81f0fc3486d49bc15a6f27c958be648cd8bb92e562217d8648a20a
+size 9907202435

latest

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

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7678721ff5facbfbfc9fa1b5cfd85099e1721d5044d227aeecb81f5a090382c
+size 4991091456

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:911f5e8c95dc954aa46fcddedb9e7f761f0c55d37a1aca317ec47c2714ef08c3
+size 1582283096

processor_config.json

@@ -0,0 +1,508 @@
+{
+  "processor_class": "Gr00tN1d6Processor",
+  "processor_kwargs": {
+    "modality_configs": {
+      "behavior_r1_pro": {
+        "video": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "observation.images.rgb.head_256_256",
+            "observation.images.rgb.left_wrist_256_256",
+            "observation.images.rgb.right_wrist_256_256"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "state": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "robot_pos",
+            "robot_ori_cos",
+            "robot_ori_sin",
+            "robot_2d_ori",
+            "robot_2d_ori_cos",
+            "robot_2d_ori_sin",
+            "robot_lin_vel",
+            "robot_ang_vel",
+            "arm_left_qpos",
+            "arm_left_qpos_sin",
+            "arm_left_qpos_cos",
+            "eef_left_pos",
+            "eef_left_quat",
+            "gripper_left_qpos",
+            "arm_right_qpos",
+            "arm_right_qpos_sin",
+            "arm_right_qpos_cos",
+            "eef_right_pos",
+            "eef_right_quat",
+            "gripper_right_qpos",
+            "trunk_qpos"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "action": {
+          "delta_indices": [
+            0,
+            1,
+            2,
+            3,
+            4,
+            5,
+            6,
+            7,
+            8,
+            9,
+            10,
+            11,
+            12,
+            13,
+            14,
+            15,
+            16,
+            17,
+            18,
+            19,
+            20,
+            21,
+            22,
+            23,
+            24,
+            25,
+            26,
+            27,
+            28,
+            29,
+            30,
+            31
+          ],
+          "modality_keys": [
+            "base",
+            "torso",
+            "left_arm",
+            "left_gripper",
+            "right_arm",
+            "right_gripper"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": [
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": "trunk_qpos"
+            },
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": "arm_left_qpos"
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": "arm_right_qpos"
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            }
+          ]
+        },
+        "language": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "annotation.human.coarse_action"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        }
+      },
+      "gr1": {
+        "video": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "ego_view_bg_crop_pad_res256_freq20"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "state": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "left_arm",
+            "right_arm",
+            "left_hand",
+            "right_hand",
+            "waist"
+          ],
+          "sin_cos_embedding_keys": [
+            "left_arm",
+            "right_arm",
+            "left_hand",
+            "right_hand",
+            "waist"
+          ],
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "action": {
+          "delta_indices": [
+            0,
+            1,
+            2,
+            3,
+            4,
+            5,
+            6,
+            7,
+            8,
+            9,
+            10,
+            11,
+            12,
+            13,
+            14,
+            15
+          ],
+          "modality_keys": [
+            "left_arm",
+            "right_arm",
+            "left_hand",
+            "right_hand",
+            "waist"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": [
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            }
+          ]
+        },
+        "language": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "task"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        }
+      },
+      "robocasa_panda_omron": {
+        "video": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "res256_image_side_0",
+            "res256_image_side_1",
+            "res256_image_wrist_0"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "state": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "end_effector_position_relative",
+            "end_effector_rotation_relative",
+            "gripper_qpos",
+            "base_position",
+            "base_rotation"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "action": {
+          "delta_indices": [
+            0,
+            1,
+            2,
+            3,
+            4,
+            5,
+            6,
+            7,
+            8,
+            9,
+            10,
+            11,
+            12,
+            13,
+            14,
+            15
+          ],
+          "modality_keys": [
+            "end_effector_position",
+            "end_effector_rotation",
+            "gripper_close",
+            "base_motion",
+            "control_mode"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": [
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            }
+          ]
+        },
+        "language": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "annotation.human.action.task_description"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        }
+      },
+      "unitree_g1": {
+        "video": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "ego_view"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "state": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "left_leg",
+            "right_leg",
+            "waist",
+            "left_arm",
+            "right_arm",
+            "left_hand",
+            "right_hand"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        },
+        "action": {
+          "delta_indices": [
+            0,
+            1,
+            2,
+            3,
+            4,
+            5,
+            6,
+            7,
+            8,
+            9,
+            10,
+            11,
+            12,
+            13,
+            14,
+            15,
+            16,
+            17,
+            18,
+            19,
+            20,
+            21,
+            22,
+            23,
+            24,
+            25,
+            26,
+            27,
+            28,
+            29
+          ],
+          "modality_keys": [
+            "left_arm",
+            "right_arm",
+            "left_hand",
+            "right_hand",
+            "waist",
+            "base_height_command",
+            "navigate_command"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": [
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "RELATIVE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            },
+            {
+              "rep": "ABSOLUTE",
+              "type": "NON_EEF",
+              "format": "DEFAULT",
+              "state_key": null
+            }
+          ]
+        },
+        "language": {
+          "delta_indices": [
+            0
+          ],
+          "modality_keys": [
+            "annotation.human.task_description"
+          ],
+          "sin_cos_embedding_keys": null,
+          "mean_std_embedding_keys": null,
+          "action_configs": null
+        }
+      }
+    },
+    "image_crop_size": null,
+    "image_target_size": null,
+    "use_albumentations": true,
+    "random_rotation_angle": null,
+    "color_jitter_params": {
+      "brightness": 0.3,
+      "contrast": 0.4,
+      "saturation": 0.5,
+      "hue": 0.08
+    },
+    "shortest_image_edge": 256,
+    "crop_fraction": 0.95,
+    "model_name": "nvidia/Eagle-Block2A-2B-v2",
+    "model_type": "eagle",
+    "formalize_language": true,
+    "max_state_dim": 128,
+    "max_action_dim": 128,
+    "max_action_horizon": 50,
+    "use_percentiles": false,
+    "clip_outliers": true,
+    "apply_sincos_state_encoding": true,
+    "use_relative_action": true
+  }
+}

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:932e1f8d91fe7ccf48c0e96843dd3b1fd68daa7444f02a740cc61e36cfb7b9db
+size 16389

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af0ec08538c0f4709cc3e28baed7e0a76a36c1222cbcb12e107b6cda14ee7b2e
+size 16389

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:779952c995e85a3f009b0cf8c2a569a944b493599a214175acbd6b1e6aac1e3b
+size 16389

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64291057621487199da0626cffc9d8f60b8ec5d0c1936cda1a8ca1a3f5725972
+size 16389

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b26f238cb7391a1f70a754fda7ae00858fca668fc362c26a4ffd5ee844dad4c8
+size 16389

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c55d9af5147ece28cc96a998fcf57a65aa2ffbb3c02d9ab7f704d7c27f2b1b52
+size 16389

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b17b143f2cd01bcee3fbf53e9d6bf20301f7a699d091277b21bab7b8f46d28d
+size 16389

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:415046c91afbb1f92fa5bf5a32ca45188793a330bf390fc5904c3bfebdd07435
+size 16389

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46c2a89c5d2cbcb36d487414dffb5e48474a9a9308a1e4dc29d920720acffd00
+size 1465

trainer_state.json

@@ -0,0 +1,1834 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6,
+  "eval_steps": 500,
+  "global_step": 3000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "grad_norm": 0.5559354424476624,
+      "learning_rate": 3.6e-06,
+      "loss": 1.0823,
+      "step": 10
+    },
+    {
+      "grad_norm": 0.18485906720161438,
+      "learning_rate": 7.6e-06,
+      "loss": 1.069,
+      "step": 20
+    },
+    {
+      "grad_norm": 0.08647036552429199,
+      "learning_rate": 1.16e-05,
+      "loss": 1.0486,
+      "step": 30
+    },
+    {
+      "grad_norm": 0.07257135212421417,
+      "learning_rate": 1.56e-05,
+      "loss": 1.0473,
+      "step": 40
+    },
+    {
+      "grad_norm": 0.08415159583091736,
+      "learning_rate": 1.9600000000000002e-05,
+      "loss": 1.043,
+      "step": 50
+    },
+    {
+      "grad_norm": 0.06655513495206833,
+      "learning_rate": 2.36e-05,
+      "loss": 1.0373,
+      "step": 60
+    },
+    {
+      "grad_norm": 0.0863005742430687,
+      "learning_rate": 2.7600000000000003e-05,
+      "loss": 1.029,
+      "step": 70
+    },
+    {
+      "grad_norm": 0.12436088174581528,
+      "learning_rate": 3.16e-05,
+      "loss": 1.0279,
+      "step": 80
+    },
+    {
+      "grad_norm": 0.6764670014381409,
+      "learning_rate": 3.56e-05,
+      "loss": 1.0159,
+      "step": 90
+    },
+    {
+      "grad_norm": 0.2262982428073883,
+      "learning_rate": 3.960000000000001e-05,
+      "loss": 0.9873,
+      "step": 100
+    },
+    {
+      "grad_norm": 0.5042984485626221,
+      "learning_rate": 4.36e-05,
+      "loss": 0.9377,
+      "step": 110
+    },
+    {
+      "grad_norm": 0.6563258767127991,
+      "learning_rate": 4.76e-05,
+      "loss": 0.849,
+      "step": 120
+    },
+    {
+      "grad_norm": 0.7196201086044312,
+      "learning_rate": 5.16e-05,
+      "loss": 0.7659,
+      "step": 130
+    },
+    {
+      "grad_norm": 0.8339490294456482,
+      "learning_rate": 5.560000000000001e-05,
+      "loss": 0.6787,
+      "step": 140
+    },
+    {
+      "grad_norm": 0.6760191321372986,
+      "learning_rate": 5.96e-05,
+      "loss": 0.6034,
+      "step": 150
+    },
+    {
+      "grad_norm": 1.257449746131897,
+      "learning_rate": 6.36e-05,
+      "loss": 0.5337,
+      "step": 160
+    },
+    {
+      "grad_norm": 0.8350836634635925,
+      "learning_rate": 6.76e-05,
+      "loss": 0.4704,
+      "step": 170
+    },
+    {
+      "grad_norm": 0.7451810836791992,
+      "learning_rate": 7.16e-05,
+      "loss": 0.4174,
+      "step": 180
+    },
+    {
+      "grad_norm": 1.0491914749145508,
+      "learning_rate": 7.560000000000001e-05,
+      "loss": 0.378,
+      "step": 190
+    },
+    {
+      "grad_norm": 0.8344352841377258,
+      "learning_rate": 7.960000000000001e-05,
+      "loss": 0.3493,
+      "step": 200
+    },
+    {
+      "grad_norm": 1.2254072427749634,
+      "learning_rate": 8.36e-05,
+      "loss": 0.3153,
+      "step": 210
+    },
+    {
+      "grad_norm": 0.7592558860778809,
+      "learning_rate": 8.76e-05,
+      "loss": 0.2874,
+      "step": 220
+    },
+    {
+      "grad_norm": 1.0830541849136353,
+      "learning_rate": 9.16e-05,
+      "loss": 0.2624,
+      "step": 230
+    },
+    {
+      "grad_norm": 0.8931087255477905,
+      "learning_rate": 9.56e-05,
+      "loss": 0.237,
+      "step": 240
+    },
+    {
+      "grad_norm": 1.8609894514083862,
+      "learning_rate": 9.960000000000001e-05,
+      "loss": 0.2146,
+      "step": 250
+    },
+    {
+      "grad_norm": 0.9318896532058716,
+      "learning_rate": 9.999911419878559e-05,
+      "loss": 0.1964,
+      "step": 260
+    },
+    {
+      "grad_norm": 0.7143105864524841,
+      "learning_rate": 9.999605221019081e-05,
+      "loss": 0.1792,
+      "step": 270
+    },
+    {
+      "grad_norm": 0.6846148371696472,
+      "learning_rate": 9.999080323230761e-05,
+      "loss": 0.1651,
+      "step": 280
+    },
+    {
+      "grad_norm": 0.9225878119468689,
+      "learning_rate": 9.998336749474329e-05,
+      "loss": 0.162,
+      "step": 290
+    },
+    {
+      "grad_norm": 0.7997626662254333,
+      "learning_rate": 9.997374532276107e-05,
+      "loss": 0.151,
+      "step": 300
+    },
+    {
+      "grad_norm": 1.0227575302124023,
+      "learning_rate": 9.996193713726596e-05,
+      "loss": 0.1404,
+      "step": 310
+    },
+    {
+      "grad_norm": 0.87727952003479,
+      "learning_rate": 9.994794345478624e-05,
+      "loss": 0.1291,
+      "step": 320
+    },
+    {
+      "grad_norm": 0.6596595644950867,
+      "learning_rate": 9.99317648874509e-05,
+      "loss": 0.1247,
+      "step": 330
+    },
+    {
+      "grad_norm": 0.8926133513450623,
+      "learning_rate": 9.991340214296292e-05,
+      "loss": 0.1181,
+      "step": 340
+    },
+    {
+      "grad_norm": 1.250920295715332,
+      "learning_rate": 9.989285602456819e-05,
+      "loss": 0.1078,
+      "step": 350
+    },
+    {
+      "grad_norm": 0.8213496804237366,
+      "learning_rate": 9.98701274310205e-05,
+      "loss": 0.1036,
+      "step": 360
+    },
+    {
+      "grad_norm": 0.8651483058929443,
+      "learning_rate": 9.984521735654218e-05,
+      "loss": 0.0871,
+      "step": 370
+    },
+    {
+      "grad_norm": 0.8947259783744812,
+      "learning_rate": 9.981812689078057e-05,
+      "loss": 0.0835,
+      "step": 380
+    },
+    {
+      "grad_norm": 0.7791630625724792,
+      "learning_rate": 9.978885721876041e-05,
+      "loss": 0.0711,
+      "step": 390
+    },
+    {
+      "grad_norm": 0.6401210427284241,
+      "learning_rate": 9.975740962083198e-05,
+      "loss": 0.0685,
+      "step": 400
+    },
+    {
+      "grad_norm": 0.7891541123390198,
+      "learning_rate": 9.972378547261504e-05,
+      "loss": 0.0649,
+      "step": 410
+    },
+    {
+      "grad_norm": 0.8624171018600464,
+      "learning_rate": 9.968798624493885e-05,
+      "loss": 0.0608,
+      "step": 420
+    },
+    {
+      "grad_norm": 0.747508704662323,
+      "learning_rate": 9.965001350377753e-05,
+      "loss": 0.0655,
+      "step": 430
+    },
+    {
+      "grad_norm": 0.9979900121688843,
+      "learning_rate": 9.960986891018183e-05,
+      "loss": 0.061,
+      "step": 440
+    },
+    {
+      "grad_norm": 0.8667426705360413,
+      "learning_rate": 9.95675542202063e-05,
+      "loss": 0.0564,
+      "step": 450
+    },
+    {
+      "grad_norm": 0.794110894203186,
+      "learning_rate": 9.952307128483256e-05,
+      "loss": 0.0534,
+      "step": 460
+    },
+    {
+      "grad_norm": 0.9227981567382812,
+      "learning_rate": 9.947642204988835e-05,
+      "loss": 0.0536,
+      "step": 470
+    },
+    {
+      "grad_norm": 0.8375693559646606,
+      "learning_rate": 9.942760855596226e-05,
+      "loss": 0.0572,
+      "step": 480
+    },
+    {
+      "grad_norm": 0.7793428897857666,
+      "learning_rate": 9.937663293831471e-05,
+      "loss": 0.0567,
+      "step": 490
+    },
+    {
+      "grad_norm": 0.91439288854599,
+      "learning_rate": 9.932349742678433e-05,
+      "loss": 0.0589,
+      "step": 500
+    },
+    {
+      "grad_norm": 0.6325005888938904,
+      "learning_rate": 9.926820434569051e-05,
+      "loss": 0.054,
+      "step": 510
+    },
+    {
+      "grad_norm": 0.7945265769958496,
+      "learning_rate": 9.921075611373179e-05,
+      "loss": 0.0572,
+      "step": 520
+    },
+    {
+      "grad_norm": 0.6466599702835083,
+      "learning_rate": 9.915115524387988e-05,
+      "loss": 0.0514,
+      "step": 530
+    },
+    {
+      "grad_norm": 0.7136776447296143,
+      "learning_rate": 9.908940434326997e-05,
+      "loss": 0.0494,
+      "step": 540
+    },
+    {
+      "grad_norm": 0.6695866584777832,
+      "learning_rate": 9.902550611308645e-05,
+      "loss": 0.0508,
+      "step": 550
+    },
+    {
+      "grad_norm": 0.7268887758255005,
+      "learning_rate": 9.895946334844494e-05,
+      "loss": 0.0488,
+      "step": 560
+    },
+    {
+      "grad_norm": 0.6591328382492065,
+      "learning_rate": 9.889127893826989e-05,
+      "loss": 0.0523,
+      "step": 570
+    },
+    {
+      "grad_norm": 0.6082274913787842,
+      "learning_rate": 9.882095586516831e-05,
+      "loss": 0.0511,
+      "step": 580
+    },
+    {
+      "grad_norm": 0.6265851855278015,
+      "learning_rate": 9.874849720529921e-05,
+      "loss": 0.0495,
+      "step": 590
+    },
+    {
+      "grad_norm": 0.6399803161621094,
+      "learning_rate": 9.867390612823914e-05,
+      "loss": 0.0508,
+      "step": 600
+    },
+    {
+      "grad_norm": 0.69126296043396,
+      "learning_rate": 9.859718589684344e-05,
+      "loss": 0.045,
+      "step": 610
+    },
+    {
+      "grad_norm": 0.6693220734596252,
+      "learning_rate": 9.851833986710353e-05,
+      "loss": 0.0479,
+      "step": 620
+    },
+    {
+      "grad_norm": 0.6840260624885559,
+      "learning_rate": 9.843737148800023e-05,
+      "loss": 0.0455,
+      "step": 630
+    },
+    {
+      "grad_norm": 0.6354555487632751,
+      "learning_rate": 9.835428430135271e-05,
+      "loss": 0.0491,
+      "step": 640
+    },
+    {
+      "grad_norm": 0.6883493661880493,
+      "learning_rate": 9.82690819416637e-05,
+      "loss": 0.047,
+      "step": 650
+    },
+    {
+      "grad_norm": 0.5793970823287964,
+      "learning_rate": 9.818176813596041e-05,
+      "loss": 0.0472,
+      "step": 660
+    },
+    {
+      "grad_norm": 0.6782571077346802,
+      "learning_rate": 9.809234670363159e-05,
+      "loss": 0.0417,
+      "step": 670
+    },
+    {
+      "grad_norm": 0.6450523734092712,
+      "learning_rate": 9.800082155626034e-05,
+      "loss": 0.048,
+      "step": 680
+    },
+    {
+      "grad_norm": 0.8551111221313477,
+      "learning_rate": 9.790719669745312e-05,
+      "loss": 0.0427,
+      "step": 690
+    },
+    {
+      "grad_norm": 0.664426863193512,
+      "learning_rate": 9.781147622266455e-05,
+      "loss": 0.0411,
+      "step": 700
+    },
+    {
+      "grad_norm": 0.625340461730957,
+      "learning_rate": 9.771366431901831e-05,
+      "loss": 0.0407,
+      "step": 710
+    },
+    {
+      "grad_norm": 0.7012608647346497,
+      "learning_rate": 9.761376526512394e-05,
+      "loss": 0.0453,
+      "step": 720
+    },
+    {
+      "grad_norm": 0.5774393677711487,
+      "learning_rate": 9.751178343088963e-05,
+      "loss": 0.0414,
+      "step": 730
+    },
+    {
+      "grad_norm": 0.6276810765266418,
+      "learning_rate": 9.740772327733123e-05,
+      "loss": 0.0393,
+      "step": 740
+    },
+    {
+      "grad_norm": 0.6698007583618164,
+      "learning_rate": 9.730158935637697e-05,
+      "loss": 0.0418,
+      "step": 750
+    },
+    {
+      "grad_norm": 0.7352779507637024,
+      "learning_rate": 9.719338631066834e-05,
+      "loss": 0.0428,
+      "step": 760
+    },
+    {
+      "grad_norm": 0.6087306141853333,
+      "learning_rate": 9.708311887335713e-05,
+      "loss": 0.0416,
+      "step": 770
+    },
+    {
+      "grad_norm": 0.7178246378898621,
+      "learning_rate": 9.697079186789823e-05,
+      "loss": 0.0404,
+      "step": 780
+    },
+    {
+      "grad_norm": 0.6541343927383423,
+      "learning_rate": 9.685641020783876e-05,
+      "loss": 0.0416,
+      "step": 790
+    },
+    {
+      "grad_norm": 0.7868788838386536,
+      "learning_rate": 9.67399788966031e-05,
+      "loss": 0.0419,
+      "step": 800
+    },
+    {
+      "grad_norm": 0.6893565058708191,
+      "learning_rate": 9.662150302727395e-05,
+      "loss": 0.0384,
+      "step": 810
+    },
+    {
+      "grad_norm": 0.555400013923645,
+      "learning_rate": 9.650098778236968e-05,
+      "loss": 0.0416,
+      "step": 820
+    },
+    {
+      "grad_norm": 0.6447874307632446,
+      "learning_rate": 9.637843843361749e-05,
+      "loss": 0.0373,
+      "step": 830
+    },
+    {
+      "grad_norm": 0.536383867263794,
+      "learning_rate": 9.62538603417229e-05,
+      "loss": 0.0396,
+      "step": 840
+    },
+    {
+      "grad_norm": 0.7303484082221985,
+      "learning_rate": 9.612725895613526e-05,
+      "loss": 0.0408,
+      "step": 850
+    },
+    {
+      "grad_norm": 0.5790427923202515,
+      "learning_rate": 9.599863981480926e-05,
+      "loss": 0.0388,
+      "step": 860
+    },
+    {
+      "grad_norm": 0.5682157874107361,
+      "learning_rate": 9.586800854396283e-05,
+      "loss": 0.0374,
+      "step": 870
+    },
+    {
+      "grad_norm": 0.6473719477653503,
+      "learning_rate": 9.573537085783095e-05,
+      "loss": 0.0362,
+      "step": 880
+    },
+    {
+      "grad_norm": 0.6345632672309875,
+      "learning_rate": 9.560073255841571e-05,
+      "loss": 0.0424,
+      "step": 890
+    },
+    {
+      "grad_norm": 0.7409985661506653,
+      "learning_rate": 9.546409953523247e-05,
+      "loss": 0.0402,
+      "step": 900
+    },
+    {
+      "grad_norm": 0.5863397121429443,
+      "learning_rate": 9.532547776505229e-05,
+      "loss": 0.0383,
+      "step": 910
+    },
+    {
+      "grad_norm": 0.5814641118049622,
+      "learning_rate": 9.518487331164048e-05,
+      "loss": 0.0365,
+      "step": 920
+    },
+    {
+      "grad_norm": 0.6481305956840515,
+      "learning_rate": 9.504229232549134e-05,
+      "loss": 0.0366,
+      "step": 930
+    },
+    {
+      "grad_norm": 0.5147769451141357,
+      "learning_rate": 9.489774104355909e-05,
+      "loss": 0.0366,
+      "step": 940
+    },
+    {
+      "grad_norm": 0.5624653100967407,
+      "learning_rate": 9.475122578898507e-05,
+      "loss": 0.0366,
+      "step": 950
+    },
+    {
+      "grad_norm": 0.6372323632240295,
+      "learning_rate": 9.460275297082119e-05,
+      "loss": 0.0352,
+      "step": 960
+    },
+    {
+      "grad_norm": 0.7020699381828308,
+      "learning_rate": 9.445232908374948e-05,
+      "loss": 0.0393,
+      "step": 970
+    },
+    {
+      "grad_norm": 0.6161628365516663,
+      "learning_rate": 9.429996070779808e-05,
+      "loss": 0.0355,
+      "step": 980
+    },
+    {
+      "grad_norm": 0.603900134563446,
+      "learning_rate": 9.414565450805333e-05,
+      "loss": 0.0383,
+      "step": 990
+    },
+    {
+      "grad_norm": 0.6112933158874512,
+      "learning_rate": 9.398941723436831e-05,
+      "loss": 0.035,
+      "step": 1000
+    },
+    {
+      "grad_norm": 0.7074472904205322,
+      "learning_rate": 9.383125572106752e-05,
+      "loss": 0.0373,
+      "step": 1010
+    },
+    {
+      "grad_norm": 0.494828462600708,
+      "learning_rate": 9.367117688664791e-05,
+      "loss": 0.0351,
+      "step": 1020
+    },
+    {
+      "grad_norm": 0.5672491788864136,
+      "learning_rate": 9.35091877334763e-05,
+      "loss": 0.0372,
+      "step": 1030
+    },
+    {
+      "grad_norm": 0.6434159874916077,
+      "learning_rate": 9.334529534748297e-05,
+      "loss": 0.0362,
+      "step": 1040
+    },
+    {
+      "grad_norm": 0.6190149784088135,
+      "learning_rate": 9.317950689785188e-05,
+      "loss": 0.0374,
+      "step": 1050
+    },
+    {
+      "grad_norm": 0.603152334690094,
+      "learning_rate": 9.301182963670688e-05,
+      "loss": 0.0363,
+      "step": 1060
+    },
+    {
+      "grad_norm": 0.661074161529541,
+      "learning_rate": 9.284227089879456e-05,
+      "loss": 0.0379,
+      "step": 1070
+    },
+    {
+      "grad_norm": 0.5567871928215027,
+      "learning_rate": 9.26708381011634e-05,
+      "loss": 0.0333,
+      "step": 1080
+    },
+    {
+      "grad_norm": 0.5913634300231934,
+      "learning_rate": 9.249753874283937e-05,
+      "loss": 0.0368,
+      "step": 1090
+    },
+    {
+      "grad_norm": 0.502541184425354,
+      "learning_rate": 9.232238040449779e-05,
+      "loss": 0.0328,
+      "step": 1100
+    },
+    {
+      "grad_norm": 0.5475773215293884,
+      "learning_rate": 9.214537074813181e-05,
+      "loss": 0.0332,
+      "step": 1110
+    },
+    {
+      "grad_norm": 0.48871368169784546,
+      "learning_rate": 9.196651751671724e-05,
+      "loss": 0.0345,
+      "step": 1120
+    },
+    {
+      "grad_norm": 0.5380433201789856,
+      "learning_rate": 9.178582853387384e-05,
+      "loss": 0.0399,
+      "step": 1130
+    },
+    {
+      "grad_norm": 0.5141293406486511,
+      "learning_rate": 9.160331170352304e-05,
+      "loss": 0.036,
+      "step": 1140
+    },
+    {
+      "grad_norm": 0.5303184986114502,
+      "learning_rate": 9.141897500954229e-05,
+      "loss": 0.0328,
+      "step": 1150
+    },
+    {
+      "grad_norm": 0.7091088891029358,
+      "learning_rate": 9.123282651541576e-05,
+      "loss": 0.035,
+      "step": 1160
+    },
+    {
+      "grad_norm": 0.5543999671936035,
+      "learning_rate": 9.104487436388161e-05,
+      "loss": 0.0364,
+      "step": 1170
+    },
+    {
+      "grad_norm": 0.5627062916755676,
+      "learning_rate": 9.085512677657582e-05,
+      "loss": 0.0347,
+      "step": 1180
+    },
+    {
+      "grad_norm": 0.5156577825546265,
+      "learning_rate": 9.066359205367258e-05,
+      "loss": 0.0346,
+      "step": 1190
+    },
+    {
+      "grad_norm": 0.49538666009902954,
+      "learning_rate": 9.047027857352112e-05,
+      "loss": 0.034,
+      "step": 1200
+    },
+    {
+      "grad_norm": 0.5321311950683594,
+      "learning_rate": 9.027519479227935e-05,
+      "loss": 0.0318,
+      "step": 1210
+    },
+    {
+      "grad_norm": 0.5262922048568726,
+      "learning_rate": 9.007834924354383e-05,
+      "loss": 0.0334,
+      "step": 1220
+    },
+    {
+      "grad_norm": 0.5841954946517944,
+      "learning_rate": 8.987975053797655e-05,
+      "loss": 0.035,
+      "step": 1230
+    },
+    {
+      "grad_norm": 0.6468002200126648,
+      "learning_rate": 8.967940736292825e-05,
+      "loss": 0.0345,
+      "step": 1240
+    },
+    {
+      "grad_norm": 0.46627357602119446,
+      "learning_rate": 8.947732848205846e-05,
+      "loss": 0.0337,
+      "step": 1250
+    },
+    {
+      "grad_norm": 0.5673134326934814,
+      "learning_rate": 8.927352273495204e-05,
+      "loss": 0.0344,
+      "step": 1260
+    },
+    {
+      "grad_norm": 0.4791153371334076,
+      "learning_rate": 8.906799903673265e-05,
+      "loss": 0.0313,
+      "step": 1270
+    },
+    {
+      "grad_norm": 0.5575198531150818,
+      "learning_rate": 8.88607663776726e-05,
+      "loss": 0.0339,
+      "step": 1280
+    },
+    {
+      "grad_norm": 0.5566645860671997,
+      "learning_rate": 8.865183382279978e-05,
+      "loss": 0.0331,
+      "step": 1290
+    },
+    {
+      "grad_norm": 0.5976455807685852,
+      "learning_rate": 8.844121051150096e-05,
+      "loss": 0.0316,
+      "step": 1300
+    },
+    {
+      "grad_norm": 0.6168678402900696,
+      "learning_rate": 8.822890565712211e-05,
+      "loss": 0.0312,
+      "step": 1310
+    },
+    {
+      "grad_norm": 0.5337743163108826,
+      "learning_rate": 8.801492854656536e-05,
+      "loss": 0.0337,
+      "step": 1320
+    },
+    {
+      "grad_norm": 0.48274630308151245,
+      "learning_rate": 8.779928853988268e-05,
+      "loss": 0.0316,
+      "step": 1330
+    },
+    {
+      "grad_norm": 0.5677248239517212,
+      "learning_rate": 8.758199506986655e-05,
+      "loss": 0.0314,
+      "step": 1340
+    },
+    {
+      "grad_norm": 0.5900443196296692,
+      "learning_rate": 8.73630576416373e-05,
+      "loss": 0.0325,
+      "step": 1350
+    },
+    {
+      "grad_norm": 0.6770350337028503,
+      "learning_rate": 8.714248583222726e-05,
+      "loss": 0.0336,
+      "step": 1360
+    },
+    {
+      "grad_norm": 0.5382674932479858,
+      "learning_rate": 8.692028929016196e-05,
+      "loss": 0.0326,
+      "step": 1370
+    },
+    {
+      "grad_norm": 0.6602237224578857,
+      "learning_rate": 8.669647773503797e-05,
+      "loss": 0.0321,
+      "step": 1380
+    },
+    {
+      "grad_norm": 0.49021854996681213,
+      "learning_rate": 8.647106095709773e-05,
+      "loss": 0.0314,
+      "step": 1390
+    },
+    {
+      "grad_norm": 0.4499266445636749,
+      "learning_rate": 8.624404881680139e-05,
+      "loss": 0.0322,
+      "step": 1400
+    },
+    {
+      "grad_norm": 0.5155117511749268,
+      "learning_rate": 8.601545124439535e-05,
+      "loss": 0.0281,
+      "step": 1410
+    },
+    {
+      "grad_norm": 0.4751155376434326,
+      "learning_rate": 8.5785278239478e-05,
+      "loss": 0.0336,
+      "step": 1420
+    },
+    {
+      "grad_norm": 0.5017455816268921,
+      "learning_rate": 8.555353987056224e-05,
+      "loss": 0.0301,
+      "step": 1430
+    },
+    {
+      "grad_norm": 0.5961615443229675,
+      "learning_rate": 8.532024627463505e-05,
+      "loss": 0.0295,
+      "step": 1440
+    },
+    {
+      "grad_norm": 0.5489777326583862,
+      "learning_rate": 8.508540765671407e-05,
+      "loss": 0.0293,
+      "step": 1450
+    },
+    {
+      "grad_norm": 0.4945319890975952,
+      "learning_rate": 8.484903428940121e-05,
+      "loss": 0.0295,
+      "step": 1460
+    },
+    {
+      "grad_norm": 0.3737506866455078,
+      "learning_rate": 8.461113651243334e-05,
+      "loss": 0.0337,
+      "step": 1470
+    },
+    {
+      "grad_norm": 0.5633870959281921,
+      "learning_rate": 8.437172473222987e-05,
+      "loss": 0.0343,
+      "step": 1480
+    },
+    {
+      "grad_norm": 0.6242406368255615,
+      "learning_rate": 8.413080942143767e-05,
+      "loss": 0.0298,
+      "step": 1490
+    },
+    {
+      "grad_norm": 0.4667608439922333,
+      "learning_rate": 8.388840111847288e-05,
+      "loss": 0.0307,
+      "step": 1500
+    },
+    {
+      "grad_norm": 0.6139534115791321,
+      "learning_rate": 8.364451042705998e-05,
+      "loss": 0.0322,
+      "step": 1510
+    },
+    {
+      "grad_norm": 0.5110672116279602,
+      "learning_rate": 8.33991480157679e-05,
+      "loss": 0.0286,
+      "step": 1520
+    },
+    {
+      "grad_norm": 0.49589312076568604,
+      "learning_rate": 8.315232461754338e-05,
+      "loss": 0.0299,
+      "step": 1530
+    },
+    {
+      "grad_norm": 0.5107879042625427,
+      "learning_rate": 8.290405102924144e-05,
+      "loss": 0.0312,
+      "step": 1540
+    },
+    {
+      "grad_norm": 0.4434208869934082,
+      "learning_rate": 8.265433811115316e-05,
+      "loss": 0.0312,
+      "step": 1550
+    },
+    {
+      "grad_norm": 0.47019222378730774,
+      "learning_rate": 8.240319678653049e-05,
+      "loss": 0.0311,
+      "step": 1560
+    },
+    {
+      "grad_norm": 0.5786542892456055,
+      "learning_rate": 8.215063804110857e-05,
+      "loss": 0.0329,
+      "step": 1570
+    },
+    {
+      "grad_norm": 0.5275363326072693,
+      "learning_rate": 8.189667292262512e-05,
+      "loss": 0.0298,
+      "step": 1580
+    },
+    {
+      "grad_norm": 0.43545565009117126,
+      "learning_rate": 8.164131254033716e-05,
+      "loss": 0.0302,
+      "step": 1590
+    },
+    {
+      "grad_norm": 0.45718058943748474,
+      "learning_rate": 8.138456806453503e-05,
+      "loss": 0.031,
+      "step": 1600
+    },
+    {
+      "grad_norm": 0.4676479697227478,
+      "learning_rate": 8.112645072605386e-05,
+      "loss": 0.029,
+      "step": 1610
+    },
+    {
+      "grad_norm": 0.5057063102722168,
+      "learning_rate": 8.086697181578222e-05,
+      "loss": 0.0327,
+      "step": 1620
+    },
+    {
+      "grad_norm": 0.47432973980903625,
+      "learning_rate": 8.060614268416823e-05,
+      "loss": 0.0311,
+      "step": 1630
+    },
+    {
+      "grad_norm": 0.44880610704421997,
+      "learning_rate": 8.034397474072309e-05,
+      "loss": 0.0288,
+      "step": 1640
+    },
+    {
+      "grad_norm": 0.42829442024230957,
+      "learning_rate": 8.008047945352193e-05,
+      "loss": 0.0289,
+      "step": 1650
+    },
+    {
+      "grad_norm": 0.4990540146827698,
+      "learning_rate": 7.981566834870225e-05,
+      "loss": 0.0316,
+      "step": 1660
+    },
+    {
+      "grad_norm": 0.4872753322124481,
+      "learning_rate": 7.954955300995961e-05,
+      "loss": 0.0295,
+      "step": 1670
+    },
+    {
+      "grad_norm": 0.4374905526638031,
+      "learning_rate": 7.928214507804104e-05,
+      "loss": 0.0295,
+      "step": 1680
+    },
+    {
+      "grad_norm": 0.4577570855617523,
+      "learning_rate": 7.901345625023576e-05,
+      "loss": 0.0279,
+      "step": 1690
+    },
+    {
+      "grad_norm": 0.570063591003418,
+      "learning_rate": 7.874349827986354e-05,
+      "loss": 0.0313,
+      "step": 1700
+    },
+    {
+      "grad_norm": 0.5689144134521484,
+      "learning_rate": 7.847228297576053e-05,
+      "loss": 0.0314,
+      "step": 1710
+    },
+    {
+      "grad_norm": 0.6005192995071411,
+      "learning_rate": 7.819982220176276e-05,
+      "loss": 0.0293,
+      "step": 1720
+    },
+    {
+      "grad_norm": 0.48968422412872314,
+      "learning_rate": 7.792612787618714e-05,
+      "loss": 0.0283,
+      "step": 1730
+    },
+    {
+      "grad_norm": 0.5264543890953064,
+      "learning_rate": 7.765121197131009e-05,
+      "loss": 0.0322,
+      "step": 1740
+    },
+    {
+      "grad_norm": 0.45399144291877747,
+      "learning_rate": 7.737508651284391e-05,
+      "loss": 0.0282,
+      "step": 1750
+    },
+    {
+      "grad_norm": 0.46420109272003174,
+      "learning_rate": 7.709776357941069e-05,
+      "loss": 0.0303,
+      "step": 1760
+    },
+    {
+      "grad_norm": 0.5926071405410767,
+      "learning_rate": 7.681925530201392e-05,
+      "loss": 0.029,
+      "step": 1770
+    },
+    {
+      "grad_norm": 0.4228203594684601,
+      "learning_rate": 7.65395738635079e-05,
+      "loss": 0.0295,
+      "step": 1780
+    },
+    {
+      "grad_norm": 0.4261937439441681,
+      "learning_rate": 7.62587314980648e-05,
+      "loss": 0.0301,
+      "step": 1790
+    },
+    {
+      "grad_norm": 0.364317923784256,
+      "learning_rate": 7.597674049063947e-05,
+      "loss": 0.0273,
+      "step": 1800
+    },
+    {
+      "grad_norm": 0.5105056166648865,
+      "learning_rate": 7.569361317643211e-05,
+      "loss": 0.0274,
+      "step": 1810
+    },
+    {
+      "grad_norm": 0.4250337779521942,
+      "learning_rate": 7.540936194034865e-05,
+      "loss": 0.0287,
+      "step": 1820
+    },
+    {
+      "grad_norm": 0.508609414100647,
+      "learning_rate": 7.512399921645901e-05,
+      "loss": 0.0263,
+      "step": 1830
+    },
+    {
+      "grad_norm": 0.43374282121658325,
+      "learning_rate": 7.483753748745317e-05,
+      "loss": 0.0271,
+      "step": 1840
+    },
+    {
+      "grad_norm": 0.5021738409996033,
+      "learning_rate": 7.454998928409516e-05,
+      "loss": 0.0285,
+      "step": 1850
+    },
+    {
+      "grad_norm": 0.43334484100341797,
+      "learning_rate": 7.426136718467493e-05,
+      "loss": 0.0271,
+      "step": 1860
+    },
+    {
+      "grad_norm": 0.6424834132194519,
+      "learning_rate": 7.397168381445812e-05,
+      "loss": 0.0289,
+      "step": 1870
+    },
+    {
+      "grad_norm": 0.4178658723831177,
+      "learning_rate": 7.368095184513377e-05,
+      "loss": 0.0296,
+      "step": 1880
+    },
+    {
+      "grad_norm": 0.37668871879577637,
+      "learning_rate": 7.338918399426005e-05,
+      "loss": 0.0262,
+      "step": 1890
+    },
+    {
+      "grad_norm": 0.35139450430870056,
+      "learning_rate": 7.309639302470801e-05,
+      "loss": 0.0273,
+      "step": 1900
+    },
+    {
+      "grad_norm": 0.5204653739929199,
+      "learning_rate": 7.280259174410312e-05,
+      "loss": 0.0314,
+      "step": 1910
+    },
+    {
+      "grad_norm": 0.44355618953704834,
+      "learning_rate": 7.250779300426517e-05,
+      "loss": 0.0266,
+      "step": 1920
+    },
+    {
+      "grad_norm": 0.4035153090953827,
+      "learning_rate": 7.22120097006461e-05,
+      "loss": 0.029,
+      "step": 1930
+    },
+    {
+      "grad_norm": 0.38775312900543213,
+      "learning_rate": 7.191525477176577e-05,
+      "loss": 0.0264,
+      "step": 1940
+    },
+    {
+      "grad_norm": 0.48700636625289917,
+      "learning_rate": 7.161754119864616e-05,
+      "loss": 0.0311,
+      "step": 1950
+    },
+    {
+      "grad_norm": 0.5373815894126892,
+      "learning_rate": 7.131888200424339e-05,
+      "loss": 0.0309,
+      "step": 1960
+    },
+    {
+      "grad_norm": 0.469199538230896,
+      "learning_rate": 7.101929025287816e-05,
+      "loss": 0.0278,
+      "step": 1970
+    },
+    {
+      "grad_norm": 0.35309985280036926,
+      "learning_rate": 7.071877904966423e-05,
+      "loss": 0.0266,
+      "step": 1980
+    },
+    {
+      "grad_norm": 0.4618619680404663,
+      "learning_rate": 7.04173615399351e-05,
+      "loss": 0.0248,
+      "step": 1990
+    },
+    {
+      "grad_norm": 0.49200907349586487,
+      "learning_rate": 7.011505090866913e-05,
+      "loss": 0.0271,
+      "step": 2000
+    },
+    {
+      "grad_norm": 0.47308212518692017,
+      "learning_rate": 6.981186037991271e-05,
+      "loss": 0.025,
+      "step": 2010
+    },
+    {
+      "grad_norm": 0.4434255361557007,
+      "learning_rate": 6.950780321620174e-05,
+      "loss": 0.0268,
+      "step": 2020
+    },
+    {
+      "grad_norm": 0.41241246461868286,
+      "learning_rate": 6.920289271798157e-05,
+      "loss": 0.0273,
+      "step": 2030
+    },
+    {
+      "grad_norm": 0.39038220047950745,
+      "learning_rate": 6.889714222302517e-05,
+      "loss": 0.0249,
+      "step": 2040
+    },
+    {
+      "grad_norm": 0.3572078049182892,
+      "learning_rate": 6.85905651058497e-05,
+      "loss": 0.0246,
+      "step": 2050
+    },
+    {
+      "grad_norm": 0.5165647864341736,
+      "learning_rate": 6.82831747771314e-05,
+      "loss": 0.0287,
+      "step": 2060
+    },
+    {
+      "grad_norm": 0.4069594740867615,
+      "learning_rate": 6.797498468311907e-05,
+      "loss": 0.027,
+      "step": 2070
+    },
+    {
+      "grad_norm": 0.5012965798377991,
+      "learning_rate": 6.766600830504585e-05,
+      "loss": 0.026,
+      "step": 2080
+    },
+    {
+      "grad_norm": 0.44795289635658264,
+      "learning_rate": 6.735625915853942e-05,
+      "loss": 0.0282,
+      "step": 2090
+    },
+    {
+      "grad_norm": 0.4666174054145813,
+      "learning_rate": 6.70457507930309e-05,
+      "loss": 0.0268,
+      "step": 2100
+    },
+    {
+      "grad_norm": 0.3589176535606384,
+      "learning_rate": 6.673449679116215e-05,
+      "loss": 0.0248,
+      "step": 2110
+    },
+    {
+      "grad_norm": 0.5180903077125549,
+      "learning_rate": 6.642251076819148e-05,
+      "loss": 0.0249,
+      "step": 2120
+    },
+    {
+      "grad_norm": 0.46328508853912354,
+      "learning_rate": 6.610980637139827e-05,
+      "loss": 0.0266,
+      "step": 2130
+    },
+    {
+      "grad_norm": 0.5663616061210632,
+      "learning_rate": 6.579639727948583e-05,
+      "loss": 0.0261,
+      "step": 2140
+    },
+    {
+      "grad_norm": 0.520371675491333,
+      "learning_rate": 6.548229720198315e-05,
+      "loss": 0.0235,
+      "step": 2150
+    },
+    {
+      "grad_norm": 0.49466976523399353,
+      "learning_rate": 6.516751987864517e-05,
+      "loss": 0.026,
+      "step": 2160
+    },
+    {
+      "grad_norm": 0.4332965314388275,
+      "learning_rate": 6.485207907885175e-05,
+      "loss": 0.0257,
+      "step": 2170
+    },
+    {
+      "grad_norm": 0.33331456780433655,
+      "learning_rate": 6.453598860100536e-05,
+      "loss": 0.0273,
+      "step": 2180
+    },
+    {
+      "grad_norm": 0.45183008909225464,
+      "learning_rate": 6.421926227192749e-05,
+      "loss": 0.0277,
+      "step": 2190
+    },
+    {
+      "grad_norm": 0.5034352540969849,
+      "learning_rate": 6.390191394625381e-05,
+      "loss": 0.0288,
+      "step": 2200
+    },
+    {
+      "grad_norm": 0.4152516722679138,
+      "learning_rate": 6.358395750582817e-05,
+      "loss": 0.0264,
+      "step": 2210
+    },
+    {
+      "grad_norm": 0.3540273606777191,
+      "learning_rate": 6.326540685909532e-05,
+      "loss": 0.0249,
+      "step": 2220
+    },
+    {
+      "grad_norm": 0.4603210389614105,
+      "learning_rate": 6.294627594049249e-05,
+      "loss": 0.0261,
+      "step": 2230
+    },
+    {
+      "grad_norm": 0.47433316707611084,
+      "learning_rate": 6.262657870983989e-05,
+      "loss": 0.0245,
+      "step": 2240
+    },
+    {
+      "grad_norm": 0.46469366550445557,
+      "learning_rate": 6.230632915173009e-05,
+      "loss": 0.0267,
+      "step": 2250
+    },
+    {
+      "grad_norm": 0.4406663775444031,
+      "learning_rate": 6.198554127491622e-05,
+      "loss": 0.0235,
+      "step": 2260
+    },
+    {
+      "grad_norm": 0.41085439920425415,
+      "learning_rate": 6.166422911169923e-05,
+      "loss": 0.0256,
+      "step": 2270
+    },
+    {
+      "grad_norm": 0.41472023725509644,
+      "learning_rate": 6.1342406717314e-05,
+      "loss": 0.0241,
+      "step": 2280
+    },
+    {
+      "grad_norm": 0.4377051889896393,
+      "learning_rate": 6.102008816931466e-05,
+      "loss": 0.0258,
+      "step": 2290
+    },
+    {
+      "grad_norm": 0.34513914585113525,
+      "learning_rate": 6.069728756695866e-05,
+      "loss": 0.0263,
+      "step": 2300
+    },
+    {
+      "grad_norm": 0.3869786858558655,
+      "learning_rate": 6.037401903059008e-05,
+      "loss": 0.026,
+      "step": 2310
+    },
+    {
+      "grad_norm": 0.3966286778450012,
+      "learning_rate": 6.005029670102195e-05,
+      "loss": 0.0247,
+      "step": 2320
+    },
+    {
+      "grad_norm": 0.35494574904441833,
+      "learning_rate": 5.972613473891766e-05,
+      "loss": 0.0256,
+      "step": 2330
+    },
+    {
+      "grad_norm": 0.31937381625175476,
+      "learning_rate": 5.940154732417158e-05,
+      "loss": 0.0231,
+      "step": 2340
+    },
+    {
+      "grad_norm": 0.38585540652275085,
+      "learning_rate": 5.907654865528876e-05,
+      "loss": 0.0277,
+      "step": 2350
+    },
+    {
+      "grad_norm": 0.3926496207714081,
+      "learning_rate": 5.875115294876381e-05,
+      "loss": 0.0238,
+      "step": 2360
+    },
+    {
+      "grad_norm": 0.3543730676174164,
+      "learning_rate": 5.842537443845908e-05,
+      "loss": 0.0236,
+      "step": 2370
+    },
+    {
+      "grad_norm": 0.36009931564331055,
+      "learning_rate": 5.809922737498198e-05,
+      "loss": 0.0251,
+      "step": 2380
+    },
+    {
+      "grad_norm": 0.3577284514904022,
+      "learning_rate": 5.777272602506165e-05,
+      "loss": 0.0264,
+      "step": 2390
+    },
+    {
+      "grad_norm": 0.4154857397079468,
+      "learning_rate": 5.744588467092483e-05,
+      "loss": 0.0261,
+      "step": 2400
+    },
+    {
+      "grad_norm": 0.3121699094772339,
+      "learning_rate": 5.7118717609671194e-05,
+      "loss": 0.0232,
+      "step": 2410
+    },
+    {
+      "grad_norm": 0.44973379373550415,
+      "learning_rate": 5.679123915264786e-05,
+      "loss": 0.0232,
+      "step": 2420
+    },
+    {
+      "grad_norm": 0.3639560639858246,
+      "learning_rate": 5.646346362482342e-05,
+      "loss": 0.023,
+      "step": 2430
+    },
+    {
+      "grad_norm": 0.5088344812393188,
+      "learning_rate": 5.613540536416132e-05,
+      "loss": 0.0256,
+      "step": 2440
+    },
+    {
+      "grad_norm": 0.45135802030563354,
+      "learning_rate": 5.5807078720992645e-05,
+      "loss": 0.0242,
+      "step": 2450
+    },
+    {
+      "grad_norm": 0.39155149459838867,
+      "learning_rate": 5.547849805738836e-05,
+      "loss": 0.0239,
+      "step": 2460
+    },
+    {
+      "grad_norm": 0.37671616673469543,
+      "learning_rate": 5.514967774653118e-05,
+      "loss": 0.0242,
+      "step": 2470
+    },
+    {
+      "grad_norm": 0.43713143467903137,
+      "learning_rate": 5.482063217208674e-05,
+      "loss": 0.0256,
+      "step": 2480
+    },
+    {
+      "grad_norm": 0.3587101995944977,
+      "learning_rate": 5.449137572757439e-05,
+      "loss": 0.0243,
+      "step": 2490
+    },
+    {
+      "grad_norm": 0.3240886926651001,
+      "learning_rate": 5.4161922815737696e-05,
+      "loss": 0.0244,
+      "step": 2500
+    },
+    {
+      "grad_norm": 0.4713996946811676,
+      "learning_rate": 5.3832287847914276e-05,
+      "loss": 0.0244,
+      "step": 2510
+    },
+    {
+      "grad_norm": 0.41371750831604004,
+      "learning_rate": 5.35024852434055e-05,
+      "loss": 0.0222,
+      "step": 2520
+    },
+    {
+      "grad_norm": 0.47091564536094666,
+      "learning_rate": 5.317252942884567e-05,
+      "loss": 0.0258,
+      "step": 2530
+    },
+    {
+      "grad_norm": 0.34738633036613464,
+      "learning_rate": 5.284243483757109e-05,
+      "loss": 0.0231,
+      "step": 2540
+    },
+    {
+      "grad_norm": 0.3713128864765167,
+      "learning_rate": 5.2512215908988484e-05,
+      "loss": 0.0239,
+      "step": 2550
+    },
+    {
+      "grad_norm": 0.41222378611564636,
+      "learning_rate": 5.218188708794357e-05,
+      "loss": 0.0233,
+      "step": 2560
+    },
+    {
+      "grad_norm": 0.3636377155780792,
+      "learning_rate": 5.18514628240891e-05,
+      "loss": 0.0232,
+      "step": 2570
+    },
+    {
+      "grad_norm": 0.3390437066555023,
+      "learning_rate": 5.1520957571252795e-05,
+      "loss": 0.0237,
+      "step": 2580
+    },
+    {
+      "grad_norm": 0.4193287491798401,
+      "learning_rate": 5.1190385786805106e-05,
+      "loss": 0.0216,
+      "step": 2590
+    },
+    {
+      "grad_norm": 0.23783208429813385,
+      "learning_rate": 5.085976193102677e-05,
+      "loss": 0.0228,
+      "step": 2600
+    },
+    {
+      "grad_norm": 0.4361535310745239,
+      "learning_rate": 5.052910046647634e-05,
+      "loss": 0.0276,
+      "step": 2610
+    },
+    {
+      "grad_norm": 0.41962987184524536,
+      "learning_rate": 5.0198415857357464e-05,
+      "loss": 0.0224,
+      "step": 2620
+    },
+    {
+      "grad_norm": 0.40996408462524414,
+      "learning_rate": 4.9867722568886223e-05,
+      "loss": 0.023,
+      "step": 2630
+    },
+    {
+      "grad_norm": 0.36495038866996765,
+      "learning_rate": 4.9537035066658314e-05,
+      "loss": 0.0234,
+      "step": 2640
+    },
+    {
+      "grad_norm": 0.42483624815940857,
+      "learning_rate": 4.920636781601638e-05,
+      "loss": 0.0256,
+      "step": 2650
+    },
+    {
+      "grad_norm": 0.39142847061157227,
+      "learning_rate": 4.88757352814172e-05,
+      "loss": 0.0289,
+      "step": 2660
+    },
+    {
+      "grad_norm": 0.3761405348777771,
+      "learning_rate": 4.8545151925798924e-05,
+      "loss": 0.0242,
+      "step": 2670
+    },
+    {
+      "grad_norm": 0.323021799325943,
+      "learning_rate": 4.821463220994848e-05,
+      "loss": 0.0227,
+      "step": 2680
+    },
+    {
+      "grad_norm": 0.4299248158931732,
+      "learning_rate": 4.788419059186895e-05,
+      "loss": 0.0218,
+      "step": 2690
+    },
+    {
+      "grad_norm": 0.3656862676143646,
+      "learning_rate": 4.7553841526147205e-05,
+      "loss": 0.0217,
+      "step": 2700
+    },
+    {
+      "grad_norm": 0.4491414725780487,
+      "learning_rate": 4.722359946332156e-05,
+      "loss": 0.0224,
+      "step": 2710
+    },
+    {
+      "grad_norm": 0.3371823728084564,
+      "learning_rate": 4.6893478849249654e-05,
+      "loss": 0.0231,
+      "step": 2720
+    },
+    {
+      "grad_norm": 0.3152349591255188,
+      "learning_rate": 4.656349412447664e-05,
+      "loss": 0.0229,
+      "step": 2730
+    },
+    {
+      "grad_norm": 0.37586328387260437,
+      "learning_rate": 4.623365972360337e-05,
+      "loss": 0.0228,
+      "step": 2740
+    },
+    {
+      "grad_norm": 0.37701380252838135,
+      "learning_rate": 4.590399007465503e-05,
+      "loss": 0.0214,
+      "step": 2750
+    },
+    {
+      "grad_norm": 0.342559814453125,
+      "learning_rate": 4.557449959845005e-05,
+      "loss": 0.0241,
+      "step": 2760
+    },
+    {
+      "grad_norm": 0.32800132036209106,
+      "learning_rate": 4.524520270796927e-05,
+      "loss": 0.0226,
+      "step": 2770
+    },
+    {
+      "grad_norm": 0.32837751507759094,
+      "learning_rate": 4.491611380772545e-05,
+      "loss": 0.0222,
+      "step": 2780
+    },
+    {
+      "grad_norm": 0.2714412808418274,
+      "learning_rate": 4.458724729313318e-05,
+      "loss": 0.0211,
+      "step": 2790
+    },
+    {
+      "grad_norm": 0.2909301221370697,
+      "learning_rate": 4.42586175498792e-05,
+      "loss": 0.0221,
+      "step": 2800
+    },
+    {
+      "grad_norm": 0.36365005373954773,
+      "learning_rate": 4.3930238953293094e-05,
+      "loss": 0.0216,
+      "step": 2810
+    },
+    {
+      "grad_norm": 0.3241809010505676,
+      "learning_rate": 4.360212586771847e-05,
+      "loss": 0.0204,
+      "step": 2820
+    },
+    {
+      "grad_norm": 0.35609936714172363,
+      "learning_rate": 4.327429264588463e-05,
+      "loss": 0.0243,
+      "step": 2830
+    },
+    {
+      "grad_norm": 0.40654465556144714,
+      "learning_rate": 4.2946753628278725e-05,
+      "loss": 0.0213,
+      "step": 2840
+    },
+    {
+      "grad_norm": 0.418736070394516,
+      "learning_rate": 4.2619523142518474e-05,
+      "loss": 0.0204,
+      "step": 2850
+    },
+    {
+      "grad_norm": 0.36981862783432007,
+      "learning_rate": 4.229261550272539e-05,
+      "loss": 0.0215,
+      "step": 2860
+    },
+    {
+      "grad_norm": 0.33936744928359985,
+      "learning_rate": 4.196604500889868e-05,
+      "loss": 0.0224,
+      "step": 2870
+    },
+    {
+      "grad_norm": 0.32824283838272095,
+      "learning_rate": 4.163982594628969e-05,
+      "loss": 0.0217,
+      "step": 2880
+    },
+    {
+      "grad_norm": 0.37890446186065674,
+      "learning_rate": 4.131397258477702e-05,
+      "loss": 0.0222,
+      "step": 2890
+    },
+    {
+      "grad_norm": 0.3364039957523346,
+      "learning_rate": 4.0988499178242315e-05,
+      "loss": 0.0223,
+      "step": 2900
+    },
+    {
+      "grad_norm": 0.4417213499546051,
+      "learning_rate": 4.066341996394678e-05,
+      "loss": 0.0209,
+      "step": 2910
+    },
+    {
+      "grad_norm": 0.4310370683670044,
+      "learning_rate": 4.033874916190833e-05,
+      "loss": 0.023,
+      "step": 2920
+    },
+    {
+      "grad_norm": 0.3380214273929596,
+      "learning_rate": 4.001450097427966e-05,
+      "loss": 0.0231,
+      "step": 2930
+    },
+    {
+      "grad_norm": 0.35001277923583984,
+      "learning_rate": 3.9690689584726894e-05,
+      "loss": 0.0234,
+      "step": 2940
+    },
+    {
+      "grad_norm": 0.412845641374588,
+      "learning_rate": 3.936732915780923e-05,
+      "loss": 0.0215,
+      "step": 2950
+    },
+    {
+      "grad_norm": 0.37060216069221497,
+      "learning_rate": 3.904443383835929e-05,
+      "loss": 0.0216,
+      "step": 2960
+    },
+    {
+      "grad_norm": 0.3685567080974579,
+      "learning_rate": 3.872201775086437e-05,
+      "loss": 0.0214,
+      "step": 2970
+    },
+    {
+      "grad_norm": 0.3236818015575409,
+      "learning_rate": 3.8400094998848616e-05,
+      "loss": 0.0198,
+      "step": 2980
+    },
+    {
+      "grad_norm": 0.2800779640674591,
+      "learning_rate": 3.807867966425611e-05,
+      "loss": 0.0219,
+      "step": 2990
+    },
+    {
+      "grad_norm": 0.2978470027446747,
+      "learning_rate": 3.775778580683481e-05,
+      "loss": 0.0231,
+      "step": 3000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 5000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 9223372036854775807,
+  "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": 0.0,
+  "train_batch_size": 96,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4bc72cc7b1b93028ee352fc4258fc426a22929f85b6310bab3ff1499a550945
+size 7633

wandb_config.json

@@ -0,0 +1 @@
+{"project": "finetune-gr00t-n1d6", "run_id": "outputs"}

zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+#   python zero_to_fp32.py . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device, weights_only=False)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+    total_files = len(files)
+    state_dicts = []
+    for f in tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info("Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info("Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)