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_block_mujoco
+    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/push_block_mujoco
+  experiment_name: null
+  max_steps: 10000
+  global_batch_size: 512
+  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: 10000
+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_block_mujoco
+    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: 512
+  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: 10000
+  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/push_block_mujoco
+  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
+}

latest

@@ -0,0 +1 @@
+global_step2000

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc4c2fb78fdbb011d0b0c5d4a62d9d5125be339f65d202319f313e69bbf21a60
+size 4991091456

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b173133b8092c8a1b87915bdbf7944ceb072c9f29bcbaddd2c1951897e8183b3
+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 sha256:0a84bc7aabbff572953927b4c15d6e1a22798b14b9c683423a5608f3e134fc36
+size 16389

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84e9786a728f346b0e876e56492ea58c046633be1f6cdc2b3b96b4bc341e157a
+size 16389

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bde98316b3f308a0d83d91c6072c60ca1f3716ecc7779efa526ac65c2387a935
+size 16389

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9152368a65056859d0730120e3bc8becff1974a2d10e4279d11699d2b563940e
+size 16389

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e15e4c8d24cfea6b4ac3ebb4af9610115799fdb244cec178e6411c8f08e5c7b
+size 16389

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3911b57f12b72079fee23232f0bdeb8b54872a7893d32a06b767fd9974fa5ed
+size 16389

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b0bf4fd46fcb118abdf3e374884f8e877873a42e23b3110394d2cb80cd2cca7
+size 16389

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8577f26da9bbddc99a57c622c90acf7149f8d84810bafec9dd09f7e3099e8e6
+size 16389

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c7bf0ea49bd9803cab03463db93f95b3262938087679ec5f7931fc0cbfeb1b1
+size 1465

trainer_state.json

@@ -0,0 +1,1234 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.2,
+  "eval_steps": 500,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "grad_norm": 1.6891276836395264,
+      "learning_rate": 1.8e-06,
+      "loss": 1.2585,
+      "step": 10
+    },
+    {
+      "grad_norm": 1.2421960830688477,
+      "learning_rate": 3.8e-06,
+      "loss": 1.2386,
+      "step": 20
+    },
+    {
+      "grad_norm": 1.0935646295547485,
+      "learning_rate": 5.8e-06,
+      "loss": 1.1761,
+      "step": 30
+    },
+    {
+      "grad_norm": 0.7210215330123901,
+      "learning_rate": 7.8e-06,
+      "loss": 1.0877,
+      "step": 40
+    },
+    {
+      "grad_norm": 0.34306979179382324,
+      "learning_rate": 9.800000000000001e-06,
+      "loss": 1.0339,
+      "step": 50
+    },
+    {
+      "grad_norm": 0.18134735524654388,
+      "learning_rate": 1.18e-05,
+      "loss": 1.0275,
+      "step": 60
+    },
+    {
+      "grad_norm": 0.15449777245521545,
+      "learning_rate": 1.3800000000000002e-05,
+      "loss": 1.0225,
+      "step": 70
+    },
+    {
+      "grad_norm": 0.15307500958442688,
+      "learning_rate": 1.58e-05,
+      "loss": 1.0188,
+      "step": 80
+    },
+    {
+      "grad_norm": 0.16959135234355927,
+      "learning_rate": 1.78e-05,
+      "loss": 1.0164,
+      "step": 90
+    },
+    {
+      "grad_norm": 0.1843302696943283,
+      "learning_rate": 1.9800000000000004e-05,
+      "loss": 1.019,
+      "step": 100
+    },
+    {
+      "grad_norm": 0.5004823803901672,
+      "learning_rate": 2.18e-05,
+      "loss": 1.007,
+      "step": 110
+    },
+    {
+      "grad_norm": 0.3533937931060791,
+      "learning_rate": 2.38e-05,
+      "loss": 0.9837,
+      "step": 120
+    },
+    {
+      "grad_norm": 0.492814302444458,
+      "learning_rate": 2.58e-05,
+      "loss": 0.95,
+      "step": 130
+    },
+    {
+      "grad_norm": 0.6054441332817078,
+      "learning_rate": 2.7800000000000005e-05,
+      "loss": 0.9005,
+      "step": 140
+    },
+    {
+      "grad_norm": 0.6737467646598816,
+      "learning_rate": 2.98e-05,
+      "loss": 0.8446,
+      "step": 150
+    },
+    {
+      "grad_norm": 1.3609671592712402,
+      "learning_rate": 3.18e-05,
+      "loss": 0.7917,
+      "step": 160
+    },
+    {
+      "grad_norm": 0.7027902603149414,
+      "learning_rate": 3.38e-05,
+      "loss": 0.7423,
+      "step": 170
+    },
+    {
+      "grad_norm": 0.6924164891242981,
+      "learning_rate": 3.58e-05,
+      "loss": 0.6849,
+      "step": 180
+    },
+    {
+      "grad_norm": 0.7687128186225891,
+      "learning_rate": 3.7800000000000004e-05,
+      "loss": 0.6385,
+      "step": 190
+    },
+    {
+      "grad_norm": 1.3768059015274048,
+      "learning_rate": 3.9800000000000005e-05,
+      "loss": 0.5934,
+      "step": 200
+    },
+    {
+      "grad_norm": 0.7467365860939026,
+      "learning_rate": 4.18e-05,
+      "loss": 0.5587,
+      "step": 210
+    },
+    {
+      "grad_norm": 0.9139049649238586,
+      "learning_rate": 4.38e-05,
+      "loss": 0.5173,
+      "step": 220
+    },
+    {
+      "grad_norm": 1.048671841621399,
+      "learning_rate": 4.58e-05,
+      "loss": 0.4742,
+      "step": 230
+    },
+    {
+      "grad_norm": 1.2771224975585938,
+      "learning_rate": 4.78e-05,
+      "loss": 0.4459,
+      "step": 240
+    },
+    {
+      "grad_norm": 1.6271154880523682,
+      "learning_rate": 4.9800000000000004e-05,
+      "loss": 0.4083,
+      "step": 250
+    },
+    {
+      "grad_norm": 1.4045976400375366,
+      "learning_rate": 5.1800000000000005e-05,
+      "loss": 0.3832,
+      "step": 260
+    },
+    {
+      "grad_norm": 1.031724214553833,
+      "learning_rate": 5.380000000000001e-05,
+      "loss": 0.3571,
+      "step": 270
+    },
+    {
+      "grad_norm": 0.9413377046585083,
+      "learning_rate": 5.580000000000001e-05,
+      "loss": 0.3294,
+      "step": 280
+    },
+    {
+      "grad_norm": 0.990053653717041,
+      "learning_rate": 5.7799999999999995e-05,
+      "loss": 0.3072,
+      "step": 290
+    },
+    {
+      "grad_norm": 0.9019013047218323,
+      "learning_rate": 5.9800000000000003e-05,
+      "loss": 0.2993,
+      "step": 300
+    },
+    {
+      "grad_norm": 0.9619300365447998,
+      "learning_rate": 6.18e-05,
+      "loss": 0.2908,
+      "step": 310
+    },
+    {
+      "grad_norm": 1.0667186975479126,
+      "learning_rate": 6.38e-05,
+      "loss": 0.2681,
+      "step": 320
+    },
+    {
+      "grad_norm": 1.119122862815857,
+      "learning_rate": 6.58e-05,
+      "loss": 0.2608,
+      "step": 330
+    },
+    {
+      "grad_norm": 1.1136410236358643,
+      "learning_rate": 6.780000000000001e-05,
+      "loss": 0.2568,
+      "step": 340
+    },
+    {
+      "grad_norm": 1.5547524690628052,
+      "learning_rate": 6.98e-05,
+      "loss": 0.2301,
+      "step": 350
+    },
+    {
+      "grad_norm": 1.0856785774230957,
+      "learning_rate": 7.18e-05,
+      "loss": 0.2166,
+      "step": 360
+    },
+    {
+      "grad_norm": 0.9325472712516785,
+      "learning_rate": 7.38e-05,
+      "loss": 0.1965,
+      "step": 370
+    },
+    {
+      "grad_norm": 0.9351710081100464,
+      "learning_rate": 7.58e-05,
+      "loss": 0.1845,
+      "step": 380
+    },
+    {
+      "grad_norm": 1.274675726890564,
+      "learning_rate": 7.780000000000001e-05,
+      "loss": 0.1701,
+      "step": 390
+    },
+    {
+      "grad_norm": 1.1908860206604004,
+      "learning_rate": 7.98e-05,
+      "loss": 0.1534,
+      "step": 400
+    },
+    {
+      "grad_norm": 1.1736817359924316,
+      "learning_rate": 8.18e-05,
+      "loss": 0.1455,
+      "step": 410
+    },
+    {
+      "grad_norm": 1.7282778024673462,
+      "learning_rate": 8.38e-05,
+      "loss": 0.1299,
+      "step": 420
+    },
+    {
+      "grad_norm": 1.0969237089157104,
+      "learning_rate": 8.58e-05,
+      "loss": 0.1252,
+      "step": 430
+    },
+    {
+      "grad_norm": 1.1998040676116943,
+      "learning_rate": 8.78e-05,
+      "loss": 0.1132,
+      "step": 440
+    },
+    {
+      "grad_norm": 1.1225069761276245,
+      "learning_rate": 8.98e-05,
+      "loss": 0.1111,
+      "step": 450
+    },
+    {
+      "grad_norm": 1.034788727760315,
+      "learning_rate": 9.180000000000001e-05,
+      "loss": 0.104,
+      "step": 460
+    },
+    {
+      "grad_norm": 1.0320745706558228,
+      "learning_rate": 9.38e-05,
+      "loss": 0.0981,
+      "step": 470
+    },
+    {
+      "grad_norm": 0.8197027444839478,
+      "learning_rate": 9.58e-05,
+      "loss": 0.0934,
+      "step": 480
+    },
+    {
+      "grad_norm": 1.0704108476638794,
+      "learning_rate": 9.78e-05,
+      "loss": 0.0844,
+      "step": 490
+    },
+    {
+      "grad_norm": 1.1035730838775635,
+      "learning_rate": 9.98e-05,
+      "loss": 0.0815,
+      "step": 500
+    },
+    {
+      "grad_norm": 0.8952515721321106,
+      "learning_rate": 9.9999778549206e-05,
+      "loss": 0.0764,
+      "step": 510
+    },
+    {
+      "grad_norm": 1.274247407913208,
+      "learning_rate": 9.999901304280685e-05,
+      "loss": 0.0774,
+      "step": 520
+    },
+    {
+      "grad_norm": 0.9264320135116577,
+      "learning_rate": 9.999770075521164e-05,
+      "loss": 0.0783,
+      "step": 530
+    },
+    {
+      "grad_norm": 0.8261769413948059,
+      "learning_rate": 9.99958417007713e-05,
+      "loss": 0.0771,
+      "step": 540
+    },
+    {
+      "grad_norm": 0.9270550608634949,
+      "learning_rate": 9.999343589981615e-05,
+      "loss": 0.0661,
+      "step": 550
+    },
+    {
+      "grad_norm": 1.1616880893707275,
+      "learning_rate": 9.999048337865568e-05,
+      "loss": 0.0755,
+      "step": 560
+    },
+    {
+      "grad_norm": 0.9049257636070251,
+      "learning_rate": 9.998698416957815e-05,
+      "loss": 0.0765,
+      "step": 570
+    },
+    {
+      "grad_norm": 1.1724106073379517,
+      "learning_rate": 9.998293831085037e-05,
+      "loss": 0.0712,
+      "step": 580
+    },
+    {
+      "grad_norm": 0.8528197407722473,
+      "learning_rate": 9.997834584671719e-05,
+      "loss": 0.0684,
+      "step": 590
+    },
+    {
+      "grad_norm": 0.9542710781097412,
+      "learning_rate": 9.997320682740107e-05,
+      "loss": 0.0716,
+      "step": 600
+    },
+    {
+      "grad_norm": 0.7609046101570129,
+      "learning_rate": 9.996752130910149e-05,
+      "loss": 0.0672,
+      "step": 610
+    },
+    {
+      "grad_norm": 0.770867109298706,
+      "learning_rate": 9.99612893539944e-05,
+      "loss": 0.0657,
+      "step": 620
+    },
+    {
+      "grad_norm": 0.5650806427001953,
+      "learning_rate": 9.995451103023144e-05,
+      "loss": 0.064,
+      "step": 630
+    },
+    {
+      "grad_norm": 0.9630634784698486,
+      "learning_rate": 9.994718641193928e-05,
+      "loss": 0.0668,
+      "step": 640
+    },
+    {
+      "grad_norm": 0.7161832451820374,
+      "learning_rate": 9.993931557921874e-05,
+      "loss": 0.0659,
+      "step": 650
+    },
+    {
+      "grad_norm": 0.9892486333847046,
+      "learning_rate": 9.993089861814402e-05,
+      "loss": 0.0627,
+      "step": 660
+    },
+    {
+      "grad_norm": 0.7624261379241943,
+      "learning_rate": 9.992193562076166e-05,
+      "loss": 0.0597,
+      "step": 670
+    },
+    {
+      "grad_norm": 0.8515845537185669,
+      "learning_rate": 9.991242668508954e-05,
+      "loss": 0.0556,
+      "step": 680
+    },
+    {
+      "grad_norm": 0.9307328462600708,
+      "learning_rate": 9.990237191511587e-05,
+      "loss": 0.053,
+      "step": 690
+    },
+    {
+      "grad_norm": 1.0245169401168823,
+      "learning_rate": 9.989177142079802e-05,
+      "loss": 0.0567,
+      "step": 700
+    },
+    {
+      "grad_norm": 0.8914056420326233,
+      "learning_rate": 9.988062531806126e-05,
+      "loss": 0.0565,
+      "step": 710
+    },
+    {
+      "grad_norm": 0.9028451442718506,
+      "learning_rate": 9.986893372879762e-05,
+      "loss": 0.0575,
+      "step": 720
+    },
+    {
+      "grad_norm": 0.8873146772384644,
+      "learning_rate": 9.985669678086443e-05,
+      "loss": 0.0602,
+      "step": 730
+    },
+    {
+      "grad_norm": 0.7956727147102356,
+      "learning_rate": 9.984391460808298e-05,
+      "loss": 0.0602,
+      "step": 740
+    },
+    {
+      "grad_norm": 0.7061982154846191,
+      "learning_rate": 9.983058735023709e-05,
+      "loss": 0.0587,
+      "step": 750
+    },
+    {
+      "grad_norm": 0.8428799510002136,
+      "learning_rate": 9.98167151530715e-05,
+      "loss": 0.0552,
+      "step": 760
+    },
+    {
+      "grad_norm": 0.9017972946166992,
+      "learning_rate": 9.980229816829034e-05,
+      "loss": 0.0595,
+      "step": 770
+    },
+    {
+      "grad_norm": 0.8163244128227234,
+      "learning_rate": 9.978733655355544e-05,
+      "loss": 0.057,
+      "step": 780
+    },
+    {
+      "grad_norm": 0.7705178260803223,
+      "learning_rate": 9.977183047248464e-05,
+      "loss": 0.0512,
+      "step": 790
+    },
+    {
+      "grad_norm": 1.0152888298034668,
+      "learning_rate": 9.975578009464992e-05,
+      "loss": 0.0532,
+      "step": 800
+    },
+    {
+      "grad_norm": 0.7115610241889954,
+      "learning_rate": 9.97391855955757e-05,
+      "loss": 0.0475,
+      "step": 810
+    },
+    {
+      "grad_norm": 0.8815870881080627,
+      "learning_rate": 9.972204715673669e-05,
+      "loss": 0.0495,
+      "step": 820
+    },
+    {
+      "grad_norm": 0.6733144521713257,
+      "learning_rate": 9.970436496555617e-05,
+      "loss": 0.05,
+      "step": 830
+    },
+    {
+      "grad_norm": 0.8259405493736267,
+      "learning_rate": 9.968613921540373e-05,
+      "loss": 0.0519,
+      "step": 840
+    },
+    {
+      "grad_norm": 0.7688018679618835,
+      "learning_rate": 9.966737010559326e-05,
+      "loss": 0.0525,
+      "step": 850
+    },
+    {
+      "grad_norm": 0.6979959607124329,
+      "learning_rate": 9.964805784138072e-05,
+      "loss": 0.0511,
+      "step": 860
+    },
+    {
+      "grad_norm": 0.7118434309959412,
+      "learning_rate": 9.962820263396195e-05,
+      "loss": 0.0482,
+      "step": 870
+    },
+    {
+      "grad_norm": 0.8046584129333496,
+      "learning_rate": 9.960780470047033e-05,
+      "loss": 0.0522,
+      "step": 880
+    },
+    {
+      "grad_norm": 0.6603678464889526,
+      "learning_rate": 9.958686426397437e-05,
+      "loss": 0.0499,
+      "step": 890
+    },
+    {
+      "grad_norm": 0.7854413986206055,
+      "learning_rate": 9.956538155347534e-05,
+      "loss": 0.052,
+      "step": 900
+    },
+    {
+      "grad_norm": 0.5703814029693604,
+      "learning_rate": 9.95433568039047e-05,
+      "loss": 0.0455,
+      "step": 910
+    },
+    {
+      "grad_norm": 0.8428476452827454,
+      "learning_rate": 9.952079025612162e-05,
+      "loss": 0.0443,
+      "step": 920
+    },
+    {
+      "grad_norm": 0.7631692886352539,
+      "learning_rate": 9.949768215691022e-05,
+      "loss": 0.0468,
+      "step": 930
+    },
+    {
+      "grad_norm": 0.7209548950195312,
+      "learning_rate": 9.9474032758977e-05,
+      "loss": 0.0474,
+      "step": 940
+    },
+    {
+      "grad_norm": 0.6969509124755859,
+      "learning_rate": 9.944984232094794e-05,
+      "loss": 0.0501,
+      "step": 950
+    },
+    {
+      "grad_norm": 0.7608206272125244,
+      "learning_rate": 9.942511110736584e-05,
+      "loss": 0.0471,
+      "step": 960
+    },
+    {
+      "grad_norm": 0.7608780860900879,
+      "learning_rate": 9.939983938868726e-05,
+      "loss": 0.0476,
+      "step": 970
+    },
+    {
+      "grad_norm": 0.6730809807777405,
+      "learning_rate": 9.93740274412797e-05,
+      "loss": 0.0431,
+      "step": 980
+    },
+    {
+      "grad_norm": 0.8944201469421387,
+      "learning_rate": 9.934767554741846e-05,
+      "loss": 0.0511,
+      "step": 990
+    },
+    {
+      "grad_norm": 0.8130436539649963,
+      "learning_rate": 9.932078399528361e-05,
+      "loss": 0.0424,
+      "step": 1000
+    },
+    {
+      "grad_norm": 0.7339826822280884,
+      "learning_rate": 9.929335307895689e-05,
+      "loss": 0.043,
+      "step": 1010
+    },
+    {
+      "grad_norm": 0.7107013463973999,
+      "learning_rate": 9.926538309841839e-05,
+      "loss": 0.0498,
+      "step": 1020
+    },
+    {
+      "grad_norm": 0.7686504125595093,
+      "learning_rate": 9.923687435954334e-05,
+      "loss": 0.0417,
+      "step": 1030
+    },
+    {
+      "grad_norm": 0.5518864989280701,
+      "learning_rate": 9.920782717409873e-05,
+      "loss": 0.0397,
+      "step": 1040
+    },
+    {
+      "grad_norm": 0.4921113848686218,
+      "learning_rate": 9.917824185973994e-05,
+      "loss": 0.0386,
+      "step": 1050
+    },
+    {
+      "grad_norm": 0.5414538979530334,
+      "learning_rate": 9.914811874000723e-05,
+      "loss": 0.0383,
+      "step": 1060
+    },
+    {
+      "grad_norm": 0.7523898482322693,
+      "learning_rate": 9.911745814432218e-05,
+      "loss": 0.0422,
+      "step": 1070
+    },
+    {
+      "grad_norm": 0.5954861640930176,
+      "learning_rate": 9.90862604079842e-05,
+      "loss": 0.0442,
+      "step": 1080
+    },
+    {
+      "grad_norm": 0.7658764719963074,
+      "learning_rate": 9.90545258721667e-05,
+      "loss": 0.0417,
+      "step": 1090
+    },
+    {
+      "grad_norm": 0.6343804001808167,
+      "learning_rate": 9.90222548839135e-05,
+      "loss": 0.0401,
+      "step": 1100
+    },
+    {
+      "grad_norm": 0.4654010534286499,
+      "learning_rate": 9.898944779613495e-05,
+      "loss": 0.0372,
+      "step": 1110
+    },
+    {
+      "grad_norm": 0.7204039096832275,
+      "learning_rate": 9.89561049676041e-05,
+      "loss": 0.0399,
+      "step": 1120
+    },
+    {
+      "grad_norm": 0.6943351626396179,
+      "learning_rate": 9.89222267629528e-05,
+      "loss": 0.0425,
+      "step": 1130
+    },
+    {
+      "grad_norm": 0.7369424104690552,
+      "learning_rate": 9.888781355266763e-05,
+      "loss": 0.0418,
+      "step": 1140
+    },
+    {
+      "grad_norm": 0.6650173664093018,
+      "learning_rate": 9.885286571308598e-05,
+      "loss": 0.0382,
+      "step": 1150
+    },
+    {
+      "grad_norm": 0.5390568375587463,
+      "learning_rate": 9.881738362639182e-05,
+      "loss": 0.0415,
+      "step": 1160
+    },
+    {
+      "grad_norm": 0.8118464946746826,
+      "learning_rate": 9.878136768061154e-05,
+      "loss": 0.0432,
+      "step": 1170
+    },
+    {
+      "grad_norm": 0.7557201981544495,
+      "learning_rate": 9.874481826960979e-05,
+      "loss": 0.0424,
+      "step": 1180
+    },
+    {
+      "grad_norm": 0.6853317022323608,
+      "learning_rate": 9.870773579308503e-05,
+      "loss": 0.0398,
+      "step": 1190
+    },
+    {
+      "grad_norm": 0.6386889815330505,
+      "learning_rate": 9.867012065656533e-05,
+      "loss": 0.0427,
+      "step": 1200
+    },
+    {
+      "grad_norm": 0.706305980682373,
+      "learning_rate": 9.863197327140376e-05,
+      "loss": 0.0375,
+      "step": 1210
+    },
+    {
+      "grad_norm": 0.7585138082504272,
+      "learning_rate": 9.859329405477403e-05,
+      "loss": 0.038,
+      "step": 1220
+    },
+    {
+      "grad_norm": 0.6497421264648438,
+      "learning_rate": 9.855408342966585e-05,
+      "loss": 0.0397,
+      "step": 1230
+    },
+    {
+      "grad_norm": 0.5648456811904907,
+      "learning_rate": 9.851434182488033e-05,
+      "loss": 0.038,
+      "step": 1240
+    },
+    {
+      "grad_norm": 0.7392773628234863,
+      "learning_rate": 9.84740696750253e-05,
+      "loss": 0.0375,
+      "step": 1250
+    },
+    {
+      "grad_norm": 0.7451732754707336,
+      "learning_rate": 9.843326742051055e-05,
+      "loss": 0.0384,
+      "step": 1260
+    },
+    {
+      "grad_norm": 0.5808190703392029,
+      "learning_rate": 9.839193550754297e-05,
+      "loss": 0.0374,
+      "step": 1270
+    },
+    {
+      "grad_norm": 0.7281758785247803,
+      "learning_rate": 9.835007438812177e-05,
+      "loss": 0.0418,
+      "step": 1280
+    },
+    {
+      "grad_norm": 0.6007897257804871,
+      "learning_rate": 9.830768452003341e-05,
+      "loss": 0.0375,
+      "step": 1290
+    },
+    {
+      "grad_norm": 0.6821262836456299,
+      "learning_rate": 9.826476636684671e-05,
+      "loss": 0.0369,
+      "step": 1300
+    },
+    {
+      "grad_norm": 0.6722646951675415,
+      "learning_rate": 9.822132039790773e-05,
+      "loss": 0.0368,
+      "step": 1310
+    },
+    {
+      "grad_norm": 0.5557746887207031,
+      "learning_rate": 9.817734708833461e-05,
+      "loss": 0.0337,
+      "step": 1320
+    },
+    {
+      "grad_norm": 0.6717876195907593,
+      "learning_rate": 9.813284691901243e-05,
+      "loss": 0.0423,
+      "step": 1330
+    },
+    {
+      "grad_norm": 0.6293877363204956,
+      "learning_rate": 9.808782037658792e-05,
+      "loss": 0.0408,
+      "step": 1340
+    },
+    {
+      "grad_norm": 0.6186095476150513,
+      "learning_rate": 9.804226795346411e-05,
+      "loss": 0.0382,
+      "step": 1350
+    },
+    {
+      "grad_norm": 0.8207311630249023,
+      "learning_rate": 9.799619014779503e-05,
+      "loss": 0.0371,
+      "step": 1360
+    },
+    {
+      "grad_norm": 0.5756433010101318,
+      "learning_rate": 9.794958746348013e-05,
+      "loss": 0.0384,
+      "step": 1370
+    },
+    {
+      "grad_norm": 0.653166651725769,
+      "learning_rate": 9.790246041015896e-05,
+      "loss": 0.034,
+      "step": 1380
+    },
+    {
+      "grad_norm": 0.6604922413825989,
+      "learning_rate": 9.785480950320538e-05,
+      "loss": 0.0347,
+      "step": 1390
+    },
+    {
+      "grad_norm": 0.7043635249137878,
+      "learning_rate": 9.78066352637221e-05,
+      "loss": 0.0338,
+      "step": 1400
+    },
+    {
+      "grad_norm": 0.6509787440299988,
+      "learning_rate": 9.775793821853488e-05,
+      "loss": 0.0355,
+      "step": 1410
+    },
+    {
+      "grad_norm": 0.6967816948890686,
+      "learning_rate": 9.77087189001868e-05,
+      "loss": 0.0353,
+      "step": 1420
+    },
+    {
+      "grad_norm": 0.8638598322868347,
+      "learning_rate": 9.765897784693243e-05,
+      "loss": 0.0361,
+      "step": 1430
+    },
+    {
+      "grad_norm": 0.6340733170509338,
+      "learning_rate": 9.760871560273197e-05,
+      "loss": 0.0351,
+      "step": 1440
+    },
+    {
+      "grad_norm": 0.6075161695480347,
+      "learning_rate": 9.755793271724526e-05,
+      "loss": 0.0342,
+      "step": 1450
+    },
+    {
+      "grad_norm": 0.6330815553665161,
+      "learning_rate": 9.750662974582584e-05,
+      "loss": 0.0406,
+      "step": 1460
+    },
+    {
+      "grad_norm": 0.5282254815101624,
+      "learning_rate": 9.745480724951473e-05,
+      "loss": 0.0351,
+      "step": 1470
+    },
+    {
+      "grad_norm": 0.7070049047470093,
+      "learning_rate": 9.740246579503447e-05,
+      "loss": 0.0366,
+      "step": 1480
+    },
+    {
+      "grad_norm": 0.7055975198745728,
+      "learning_rate": 9.734960595478284e-05,
+      "loss": 0.0357,
+      "step": 1490
+    },
+    {
+      "grad_norm": 0.6384962797164917,
+      "learning_rate": 9.729622830682657e-05,
+      "loss": 0.0343,
+      "step": 1500
+    },
+    {
+      "grad_norm": 0.5874096751213074,
+      "learning_rate": 9.724233343489504e-05,
+      "loss": 0.0372,
+      "step": 1510
+    },
+    {
+      "grad_norm": 0.5690801739692688,
+      "learning_rate": 9.718792192837396e-05,
+      "loss": 0.0319,
+      "step": 1520
+    },
+    {
+      "grad_norm": 0.6367931962013245,
+      "learning_rate": 9.713299438229886e-05,
+      "loss": 0.037,
+      "step": 1530
+    },
+    {
+      "grad_norm": 0.5469662547111511,
+      "learning_rate": 9.707755139734855e-05,
+      "loss": 0.0365,
+      "step": 1540
+    },
+    {
+      "grad_norm": 0.556199848651886,
+      "learning_rate": 9.702159357983866e-05,
+      "loss": 0.0342,
+      "step": 1550
+    },
+    {
+      "grad_norm": 0.6330277919769287,
+      "learning_rate": 9.696512154171492e-05,
+      "loss": 0.0363,
+      "step": 1560
+    },
+    {
+      "grad_norm": 0.6156578660011292,
+      "learning_rate": 9.690813590054645e-05,
+      "loss": 0.036,
+      "step": 1570
+    },
+    {
+      "grad_norm": 0.633952260017395,
+      "learning_rate": 9.685063727951914e-05,
+      "loss": 0.0342,
+      "step": 1580
+    },
+    {
+      "grad_norm": 0.4720134735107422,
+      "learning_rate": 9.679262630742865e-05,
+      "loss": 0.0352,
+      "step": 1590
+    },
+    {
+      "grad_norm": 0.6747561097145081,
+      "learning_rate": 9.673410361867373e-05,
+      "loss": 0.0361,
+      "step": 1600
+    },
+    {
+      "grad_norm": 0.5455501675605774,
+      "learning_rate": 9.667506985324909e-05,
+      "loss": 0.0332,
+      "step": 1610
+    },
+    {
+      "grad_norm": 0.698010265827179,
+      "learning_rate": 9.661552565673855e-05,
+      "loss": 0.0309,
+      "step": 1620
+    },
+    {
+      "grad_norm": 0.6324073672294617,
+      "learning_rate": 9.655547168030789e-05,
+      "loss": 0.035,
+      "step": 1630
+    },
+    {
+      "grad_norm": 0.6724660992622375,
+      "learning_rate": 9.649490858069777e-05,
+      "loss": 0.0327,
+      "step": 1640
+    },
+    {
+      "grad_norm": 0.4986797273159027,
+      "learning_rate": 9.643383702021658e-05,
+      "loss": 0.032,
+      "step": 1650
+    },
+    {
+      "grad_norm": 0.6412114500999451,
+      "learning_rate": 9.637225766673307e-05,
+      "loss": 0.0309,
+      "step": 1660
+    },
+    {
+      "grad_norm": 0.5809900164604187,
+      "learning_rate": 9.631017119366922e-05,
+      "loss": 0.0319,
+      "step": 1670
+    },
+    {
+      "grad_norm": 0.5835633277893066,
+      "learning_rate": 9.624757827999273e-05,
+      "loss": 0.0339,
+      "step": 1680
+    },
+    {
+      "grad_norm": 0.5545461773872375,
+      "learning_rate": 9.618447961020971e-05,
+      "loss": 0.039,
+      "step": 1690
+    },
+    {
+      "grad_norm": 0.6023017168045044,
+      "learning_rate": 9.612087587435707e-05,
+      "loss": 0.038,
+      "step": 1700
+    },
+    {
+      "grad_norm": 0.6422150135040283,
+      "learning_rate": 9.605676776799508e-05,
+      "loss": 0.0342,
+      "step": 1710
+    },
+    {
+      "grad_norm": 0.6100661158561707,
+      "learning_rate": 9.599215599219973e-05,
+      "loss": 0.0332,
+      "step": 1720
+    },
+    {
+      "grad_norm": 0.6025664806365967,
+      "learning_rate": 9.592704125355505e-05,
+      "loss": 0.0332,
+      "step": 1730
+    },
+    {
+      "grad_norm": 0.5365665555000305,
+      "learning_rate": 9.586142426414538e-05,
+      "loss": 0.0322,
+      "step": 1740
+    },
+    {
+      "grad_norm": 0.6416350603103638,
+      "learning_rate": 9.57953057415476e-05,
+      "loss": 0.0356,
+      "step": 1750
+    },
+    {
+      "grad_norm": 0.5091562867164612,
+      "learning_rate": 9.572868640882328e-05,
+      "loss": 0.0327,
+      "step": 1760
+    },
+    {
+      "grad_norm": 0.506897509098053,
+      "learning_rate": 9.56615669945108e-05,
+      "loss": 0.0347,
+      "step": 1770
+    },
+    {
+      "grad_norm": 0.5713956356048584,
+      "learning_rate": 9.55939482326173e-05,
+      "loss": 0.0324,
+      "step": 1780
+    },
+    {
+      "grad_norm": 0.47305065393447876,
+      "learning_rate": 9.552583086261069e-05,
+      "loss": 0.034,
+      "step": 1790
+    },
+    {
+      "grad_norm": 0.5677897334098816,
+      "learning_rate": 9.545721562941168e-05,
+      "loss": 0.0319,
+      "step": 1800
+    },
+    {
+      "grad_norm": 0.5385820865631104,
+      "learning_rate": 9.538810328338543e-05,
+      "loss": 0.0306,
+      "step": 1810
+    },
+    {
+      "grad_norm": 0.67743319272995,
+      "learning_rate": 9.531849458033349e-05,
+      "loss": 0.0336,
+      "step": 1820
+    },
+    {
+      "grad_norm": 0.6273674368858337,
+      "learning_rate": 9.524839028148547e-05,
+      "loss": 0.0325,
+      "step": 1830
+    },
+    {
+      "grad_norm": 0.5729127526283264,
+      "learning_rate": 9.517779115349077e-05,
+      "loss": 0.033,
+      "step": 1840
+    },
+    {
+      "grad_norm": 0.5738729238510132,
+      "learning_rate": 9.510669796841014e-05,
+      "loss": 0.0327,
+      "step": 1850
+    },
+    {
+      "grad_norm": 0.46439313888549805,
+      "learning_rate": 9.503511150370727e-05,
+      "loss": 0.0349,
+      "step": 1860
+    },
+    {
+      "grad_norm": 0.49889275431632996,
+      "learning_rate": 9.496303254224024e-05,
+      "loss": 0.0329,
+      "step": 1870
+    },
+    {
+      "grad_norm": 0.466863751411438,
+      "learning_rate": 9.489046187225306e-05,
+      "loss": 0.032,
+      "step": 1880
+    },
+    {
+      "grad_norm": 0.49700266122817993,
+      "learning_rate": 9.481740028736692e-05,
+      "loss": 0.0335,
+      "step": 1890
+    },
+    {
+      "grad_norm": 0.6205595135688782,
+      "learning_rate": 9.474384858657164e-05,
+      "loss": 0.0315,
+      "step": 1900
+    },
+    {
+      "grad_norm": 0.5160883665084839,
+      "learning_rate": 9.466980757421679e-05,
+      "loss": 0.0313,
+      "step": 1910
+    },
+    {
+      "grad_norm": 0.5717476010322571,
+      "learning_rate": 9.459527806000305e-05,
+      "loss": 0.0333,
+      "step": 1920
+    },
+    {
+      "grad_norm": 0.5541268587112427,
+      "learning_rate": 9.452026085897325e-05,
+      "loss": 0.0327,
+      "step": 1930
+    },
+    {
+      "grad_norm": 0.5027980208396912,
+      "learning_rate": 9.444475679150348e-05,
+      "loss": 0.0314,
+      "step": 1940
+    },
+    {
+      "grad_norm": 0.5508278608322144,
+      "learning_rate": 9.436876668329411e-05,
+      "loss": 0.0303,
+      "step": 1950
+    },
+    {
+      "grad_norm": 0.5171688199043274,
+      "learning_rate": 9.429229136536079e-05,
+      "loss": 0.0298,
+      "step": 1960
+    },
+    {
+      "grad_norm": 0.6098824143409729,
+      "learning_rate": 9.421533167402534e-05,
+      "loss": 0.0314,
+      "step": 1970
+    },
+    {
+      "grad_norm": 0.5086256265640259,
+      "learning_rate": 9.413788845090666e-05,
+      "loss": 0.0317,
+      "step": 1980
+    },
+    {
+      "grad_norm": 0.6548255681991577,
+      "learning_rate": 9.405996254291136e-05,
+      "loss": 0.0311,
+      "step": 1990
+    },
+    {
+      "grad_norm": 0.6564981341362,
+      "learning_rate": 9.398155480222474e-05,
+      "loss": 0.0309,
+      "step": 2000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 10000,
+  "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": 64,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:649db4ea79a44576dda217d318af2d5628507b2af4facdb1a441e9dbdcaf8408
+size 7633

wandb_config.json

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

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)