upload VLA-UniT-3B-fulldata checkpoint

VLA-UniT-3B-fulldata/config.json

@@ -0,0 +1,100 @@
+{
+  "action_dim": 128,
+  "action_head_cfg": {
+    "action_dim": 128,
+    "action_horizon": 16,
+    "add_pos_embed": true,
+    "backbone_embedding_dim": 2048,
+    "diffusion_model_cfg": {
+      "attention_head_dim": 48,
+      "cross_attention_dim": 2048,
+      "dropout": 0.2,
+      "final_dropout": true,
+      "interleave_self_attention": true,
+      "norm_type": "ada_norm",
+      "num_attention_heads": 32,
+      "num_layers": 16,
+      "output_dim": 1024,
+      "positional_embeddings": null
+    },
+    "hidden_size": 1024,
+    "input_embedding_dim": 1536,
+    "max_action_dim": 128,
+    "max_state_dim": 128,
+    "model_dtype": "float32",
+    "noise_beta_alpha": 1.5,
+    "noise_beta_beta": 1.0,
+    "noise_s": 0.999,
+    "num_inference_timesteps": 4,
+    "num_target_vision_tokens": 0,
+    "num_timestep_buckets": 1000,
+    "tune_diffusion_model": true,
+    "tune_projector": true,
+    "use_correct_attn_mask": true,
+    "use_vl_mask": true,
+    "use_vlln": true,
+    "vl_self_attention_cfg": {
+      "attention_head_dim": 64,
+      "dropout": 0.2,
+      "final_dropout": true,
+      "num_attention_heads": 32,
+      "num_layers": 4,
+      "positional_embeddings": null
+    }
+  },
+  "action_horizon": 16,
+  "architectures": [
+    "GR00T_N1_5_UniT"
+  ],
+  "attn_implementation": null,
+  "backbone_cfg": {
+    "eagle_path": "Qwen/Qwen2.5-VL-3B-Instruct",
+    "load_bf16": false,
+    "project_to_dim": null,
+    "reproject_vision": false,
+    "select_layer": 36,
+    "tune_all_llm_embedding": false,
+    "tune_bridge_embedding": true,
+    "tune_llm": true,
+    "tune_visual": false,
+    "use_flash_attention": true
+  },
+  "compute_dtype": "bfloat16",
+  "hidden_size": 2048,
+  "ignore_lang_prefix": true,
+  "model_dtype": "float32",
+  "model_type": "gr00t_n1_5_unit",
+  "torch_dtype": "float32",
+  "transformers_version": "4.52.0",
+  "unit_cfg": {
+    "action_only_one_obs": false,
+    "bridge_loss_type": "ce",
+    "bridge_type": "vision_lang_obs",
+    "compute_bridge_loss": true,
+    "groot_tokenizer_path": "tokenizer",
+    "noise_tau": 0,
+    "num_bridge_tokens": 8,
+    "omit_image_type_embedding_for_goal": false,
+    "reweight_noise": false,
+    "tokenizer_cfg": {
+      "ce_loss_weights": [
+        1.0,
+        0.5
+      ],
+      "codebook_dim": 32,
+      "codebook_size": 128,
+      "label_smoothing": 0.0,
+      "num_codebooks": 2,
+      "vis_freq": 100
+    },
+    "tokenizer_len": 151673,
+    "tune_bridge_visual": false,
+    "tune_image_type_embedding": true,
+    "unified_embodiment_id": null,
+    "use_bridge": true,
+    "use_image_type_embedding": true
+  },
+  "video_delta_indices": [
+    0
+  ]
+}

VLA-UniT-3B-fulldata/model-00001-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:832163366d74b012a6c3d763422ad18dbf2ad9a20487b3ed22e8e35799a80567
+size 4972313928

VLA-UniT-3B-fulldata/model-00002-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ced52fe60d5a77b935101e892312d347bb3342dbba8ec5e992e267eda301944
+size 4932955008

VLA-UniT-3B-fulldata/model-00003-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:202e7a5f34ea0cb2768caa3fe3d5e3b53cfe3077040db95af6420564d8596239
+size 4932955096

VLA-UniT-3B-fulldata/model-00004-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e946f686f10c277ae800e65d8acc71ae0786436261485c6ba57824830fad9776
+size 4531035752

VLA-UniT-3B-fulldata/tokenizer/config.json

@@ -0,0 +1,189 @@
+{
+  "_comment_action_decoder": "========== Action Decoder (ResNet-based Direct Reconstruction) ==========",
+  "_comment_action_encoder": "========== Action Encoder ==========",
+  "_comment_backbone": "========== Backbone Configuration (Vision Model Only) ==========",
+  "_comment_basic": "========== Basic Parameters ==========",
+  "_comment_debug": "========== Debug Configuration ==========",
+  "_comment_fusion": "========== Visual-Action Fusion ==========",
+  "_comment_lpips": "DINO mode: LPIPS disabled, using cosine similarity for vision reconstruction",
+  "_comment_modality": "========== Modality Presence Configuration ==========",
+  "_comment_trainable": "========== Default Trainable Flags ==========",
+  "_comment_training": "========== Training Configuration (DINO Mode) ==========",
+  "_comment_vision_branch": "========== Vision Branch Encoder ==========",
+  "_comment_vision_decoder": "========== Vision Decoder (DINO Hidden State Mode) ==========",
+  "_comment_visualization": "========== Visualization Configuration (Disabled for DINO) ==========",
+  "_comment_vq": "========== Vector Quantizer ==========",
+  "action_decoder_cfg": {
+    "_comment_decoder_type": "Decoder type: 'resnet' (direct reconstruction) or 'flow_matching' (diffusion)",
+    "_comment_embodiment": "Embodiment configuration",
+    "_comment_m_former": "M-Former configuration (symmetric to encoder)",
+    "_comment_resnet": "ResNet decoder configuration (symmetric to encoder)",
+    "action_dim": 128,
+    "action_horizon": 16,
+    "conv_kernel_size": 5,
+    "decoder_type": "resnet",
+    "dropout": 0.0,
+    "hidden_size": 1024,
+    "m_former_cfg": {
+      "attention_probs_dropout_prob": 0.0,
+      "hidden_dropout_prob": 0.0,
+      "hidden_size": 1024,
+      "input_hidden_size": 1024,
+      "intermediate_size": 3072,
+      "num_attention_heads": 16,
+      "num_hidden_layers": 8,
+      "query_num": 8
+    },
+    "max_num_embodiments": 30,
+    "num_conv_layers": 3,
+    "query_num": 8,
+    "upsample_stride": 2,
+    "use_dilation": false
+  },
+  "action_dim": 128,
+  "action_encoder_cfg": {
+    "action_dim": 128,
+    "action_horizon": 16,
+    "conv_encoder_type": "resnet",
+    "conv_kernel_size": 5,
+    "conv_stride": 2,
+    "downsample_target_len": null,
+    "dropout": 0.0,
+    "hidden_size": 1024,
+    "m_former_cfg": {
+      "attention_probs_dropout_prob": 0.0,
+      "hidden_dropout_prob": 0.0,
+      "hidden_size": 1024,
+      "input_hidden_size": 1024,
+      "intermediate_size": 3072,
+      "num_attention_heads": 16,
+      "num_hidden_layers": 4,
+      "query_num": 8
+    },
+    "max_num_embodiments": 30,
+    "num_conv_layers": 3,
+    "query_num": 8,
+    "state_dim": 128,
+    "use_dilation": true
+  },
+  "action_horizon": 16,
+  "architectures": [
+    "GR00T_Tokenizer"
+  ],
+  "attn_implementation": null,
+  "backbone_cfg": {
+    "_comment_layer_index": "Layer index: -1=last layer, -2=second-to-last, 0-N=specific layer",
+    "dinov2_layer_index": -2,
+    "dinov2_path": "facebook/dinov2-large",
+    "eagle_path": "Qwen/Qwen2.5-VL-3B-Instruct",
+    "load_bf16": false,
+    "project_to_dim": null,
+    "reproject_vision": false,
+    "select_layer": 12,
+    "tune_bridge_embedding": false,
+    "tune_llm": false,
+    "tune_visual": false,
+    "use_flash_attention": true,
+    "use_torch_compile": false,
+    "vision_model_type": "dinov2"
+  },
+  "bridge_projector_mid_dim": 1024,
+  "compute_dtype": "bfloat16",
+  "default_pa": 1,
+  "default_pv": 1,
+  "enable_action_visualization": false,
+  "enable_dimension_debug": false,
+  "enable_visualization": false,
+  "fusion_cfg": {
+    "dropout": 0.0,
+    "hidden_size": 1024,
+    "num_attention_heads": 16,
+    "num_layers": 4,
+    "query_num": 8
+  },
+  "hidden_size": 1024,
+  "ignore_lang_prefix": false,
+  "model_type": "gr00t_tokenizer",
+  "query_num": 8,
+  "reconstruction_loss_weights": {
+    "action": 1.0,
+    "lpips": 0.0,
+    "vision": 1.0,
+    "vq_commitment": 1.0
+  },
+  "state_dim": 128,
+  "torch_dtype": "float32",
+  "transformers_version": "4.52.0",
+  "tune_action_decoder_diffusion": true,
+  "tune_action_decoder_projector": true,
+  "tune_action_encoder": true,
+  "tune_bridge_projector": true,
+  "tune_fusion": true,
+  "tune_image_type_embedding": true,
+  "tune_vision_decoder": true,
+  "tune_vision_m_former": true,
+  "tune_vision_model": false,
+  "tune_vq": true,
+  "unified_embodiment_id": null,
+  "use_image_type_embedding": true,
+  "use_lpips_loss": false,
+  "use_multi_scenario_training": true,
+  "video_delta_indices": [
+    0
+  ],
+  "vision_branch_cfg": {
+    "hidden_size": 1024,
+    "m_former_cfg": {
+      "attention_probs_dropout_prob": 0.0,
+      "hidden_dropout_prob": 0.0,
+      "hidden_size": 1024,
+      "input_hidden_size": 1024,
+      "intermediate_size": 3072,
+      "max_seq_len": 1024,
+      "num_attention_heads": 16,
+      "num_hidden_layers": 4,
+      "query_num": 8
+    },
+    "query_num": 8
+  },
+  "vision_decoder_cfg": {
+    "_comment_dino": "DINO mode: input/output are hidden states, not images",
+    "attention_probs_dropout_prob": 0.0,
+    "encoder_stride": 14,
+    "hidden_dropout_prob": 0.0,
+    "hidden_size": 1024,
+    "image_size": 224,
+    "input_hidden_size": 1024,
+    "intermediate_size": 3072,
+    "is_io_hidden_states": true,
+    "latent_motion_dim": 1024,
+    "num_attention_heads": 16,
+    "num_channels": 3,
+    "num_hidden_layers": 12,
+    "num_patches": 256,
+    "patch_size": 14,
+    "query_num": 8
+  },
+  "visualization_freq": 1000,
+  "visualization_num_samples": 4,
+  "visualization_save_dir": "visualizations",
+  "vq_cfg": {
+    "beta": 0.25,
+    "e_dim": 32,
+    "n_e": 128,
+    "num_stages": 2,
+    "sane_index_shape": false,
+    "stages": [
+      {
+        "beta": 0.5,
+        "e_dim": 32,
+        "n_e": 128
+      },
+      {
+        "beta": 0.1,
+        "e_dim": 32,
+        "n_e": 128
+      }
+    ]
+  }
+}

VLA-UniT-3B-fulldata/tokenizer/model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32d5c326f6c83d12185b6954d2a52511f66ad18b6fdf814aecc5726dd39c243c
+size 4786382848

VLA-UniT-3B-fulldata/tokenizer/model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f8093a900330e5111b63e44dc1687b3212bec343e5b3832bf2e40f2bf18a768
+size 708440360