Upload policy weights, train config and readme

README.md

@@ -1,21 +1,22 @@
 ---
+base_model: lerobot/smolvla_base
 datasets: data/merged_multitask
 library_name: lerobot
 license: apache-2.0
-model_name: diffusion
+model_name: smolvla
 pipeline_tag: robotics
 tags:
 - lerobot
 - robotics
-- diffusion
+- smolvla
 ---
 
-# Model Card for diffusion
+# Model Card for smolvla
 
 <!-- Provide a quick summary of what the model is/does. -->
 
 
-[Diffusion Policy](https://huggingface.co/papers/2303.04137) treats visuomotor control as a generative diffusion process, producing smooth, multi-step action trajectories that excel at contact-rich manipulation.
+[SmolVLA](https://huggingface.co/papers/2506.01844) is a compact, efficient vision-language-action model that achieves competitive performance at reduced computational costs and can be deployed on consumer-grade hardware.
 
 
 This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).

config.json

@@ -1,6 +1,6 @@
 {
-    "type": "diffusion",
-    "n_obs_steps": 2,
+    "type": "smolvla",
+    "n_obs_steps": 1,
     "input_features": {
         "observation.state": {
             "type": "STATE",
@@ -49,50 +49,49 @@
     "tags": null,
     "license": null,
     "pretrained_path": null,
-    "horizon": 16,
-    "n_action_steps": 8,
+    "chunk_size": 50,
+    "n_action_steps": 50,
     "normalization_mapping": {
-        "VISUAL": "MEAN_STD",
-        "STATE": "MIN_MAX",
-        "ACTION": "MIN_MAX"
+        "VISUAL": "IDENTITY",
+        "STATE": "MEAN_STD",
+        "ACTION": "MEAN_STD"
     },
-    "drop_n_last_frames": 7,
-    "vision_backbone": "resnet18",
-    "crop_shape": [
-        84,
-        84
-    ],
-    "crop_is_random": true,
-    "pretrained_backbone_weights": null,
-    "use_group_norm": true,
-    "spatial_softmax_num_keypoints": 32,
-    "use_separate_rgb_encoder_per_camera": false,
-    "down_dims": [
+    "max_state_dim": 32,
+    "max_action_dim": 32,
+    "resize_imgs_with_padding": [
         512,
-        1024,
-        2048
+        512
     ],
-    "kernel_size": 5,
-    "n_groups": 8,
-    "diffusion_step_embed_dim": 128,
-    "use_film_scale_modulation": true,
-    "noise_scheduler_type": "DDPM",
-    "num_train_timesteps": 100,
-    "beta_schedule": "squaredcos_cap_v2",
-    "beta_start": 0.0001,
-    "beta_end": 0.02,
-    "prediction_type": "epsilon",
-    "clip_sample": true,
-    "clip_sample_range": 1.0,
-    "num_inference_steps": null,
-    "do_mask_loss_for_padding": false,
+    "empty_cameras": 0,
+    "adapt_to_pi_aloha": false,
+    "use_delta_joint_actions_aloha": false,
+    "tokenizer_max_length": 48,
+    "num_steps": 10,
+    "use_cache": true,
+    "freeze_vision_encoder": true,
+    "train_expert_only": true,
+    "train_state_proj": true,
     "optimizer_lr": 0.0001,
     "optimizer_betas": [
-        0.95,
-        0.999
+        0.9,
+        0.95
     ],
     "optimizer_eps": 1e-08,
-    "optimizer_weight_decay": 1e-06,
-    "scheduler_name": "cosine",
-    "scheduler_warmup_steps": 500
+    "optimizer_weight_decay": 1e-10,
+    "optimizer_grad_clip_norm": 10,
+    "scheduler_warmup_steps": 1000,
+    "scheduler_decay_steps": 30000,
+    "scheduler_decay_lr": 2.5e-06,
+    "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
+    "load_vlm_weights": false,
+    "add_image_special_tokens": false,
+    "attention_mode": "cross_attn",
+    "prefix_length": -1,
+    "pad_language_to": "longest",
+    "num_expert_layers": -1,
+    "num_vlm_layers": 16,
+    "self_attn_every_n_layers": 2,
+    "expert_width_multiplier": 0.75,
+    "min_period": 0.004,
+    "max_period": 4.0
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0104bcfa0d8809416cd4016f0653b497e5d9752cd336bfd664550879ed17e6cf
-size 1082658752
+oid sha256:8378d98d2f003917a142cf2bc75628f6a3adb04be96ef1822f149bd5436282e9
+size 1197789224

train_config.json

@@ -81,8 +81,8 @@
     },
     "env": null,
     "policy": {
-        "type": "diffusion",
-        "n_obs_steps": 2,
+        "type": "smolvla",
+        "n_obs_steps": 1,
         "input_features": {
             "observation.state": {
                 "type": "STATE",
@@ -131,55 +131,54 @@
         "tags": null,
         "license": null,
         "pretrained_path": null,
-        "horizon": 16,
-        "n_action_steps": 8,
+        "chunk_size": 50,
+        "n_action_steps": 50,
         "normalization_mapping": {
-            "VISUAL": "MEAN_STD",
-            "STATE": "MIN_MAX",
-            "ACTION": "MIN_MAX"
+            "VISUAL": "IDENTITY",
+            "STATE": "MEAN_STD",
+            "ACTION": "MEAN_STD"
         },
-        "drop_n_last_frames": 7,
-        "vision_backbone": "resnet18",
-        "crop_shape": [
-            84,
-            84
-        ],
-        "crop_is_random": true,
-        "pretrained_backbone_weights": null,
-        "use_group_norm": true,
-        "spatial_softmax_num_keypoints": 32,
-        "use_separate_rgb_encoder_per_camera": false,
-        "down_dims": [
+        "max_state_dim": 32,
+        "max_action_dim": 32,
+        "resize_imgs_with_padding": [
             512,
-            1024,
-            2048
+            512
         ],
-        "kernel_size": 5,
-        "n_groups": 8,
-        "diffusion_step_embed_dim": 128,
-        "use_film_scale_modulation": true,
-        "noise_scheduler_type": "DDPM",
-        "num_train_timesteps": 100,
-        "beta_schedule": "squaredcos_cap_v2",
-        "beta_start": 0.0001,
-        "beta_end": 0.02,
-        "prediction_type": "epsilon",
-        "clip_sample": true,
-        "clip_sample_range": 1.0,
-        "num_inference_steps": null,
-        "do_mask_loss_for_padding": false,
+        "empty_cameras": 0,
+        "adapt_to_pi_aloha": false,
+        "use_delta_joint_actions_aloha": false,
+        "tokenizer_max_length": 48,
+        "num_steps": 10,
+        "use_cache": true,
+        "freeze_vision_encoder": true,
+        "train_expert_only": true,
+        "train_state_proj": true,
         "optimizer_lr": 0.0001,
         "optimizer_betas": [
-            0.95,
-            0.999
+            0.9,
+            0.95
         ],
         "optimizer_eps": 1e-08,
-        "optimizer_weight_decay": 1e-06,
-        "scheduler_name": "cosine",
-        "scheduler_warmup_steps": 500
+        "optimizer_weight_decay": 1e-10,
+        "optimizer_grad_clip_norm": 10,
+        "scheduler_warmup_steps": 1000,
+        "scheduler_decay_steps": 30000,
+        "scheduler_decay_lr": 2.5e-06,
+        "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
+        "load_vlm_weights": false,
+        "add_image_special_tokens": false,
+        "attention_mode": "cross_attn",
+        "prefix_length": -1,
+        "pad_language_to": "longest",
+        "num_expert_layers": -1,
+        "num_vlm_layers": 16,
+        "self_attn_every_n_layers": 2,
+        "expert_width_multiplier": 0.75,
+        "min_period": 0.004,
+        "max_period": 4.0
     },
-    "output_dir": "/media/choi/HDD/outputs/train/Diffusion_merged",
-    "job_name": "diffusion_merged",
+    "output_dir": "/media/choi/HDD/outputs/train/smolvla_merged",
+    "job_name": "smolvla_merged",
     "resume": false,
     "seed": 1000,
     "num_workers": 4,
@@ -191,20 +190,22 @@
     "save_freq": 20000,
     "use_policy_training_preset": true,
     "optimizer": {
-        "type": "adam",
+        "type": "adamw",
         "lr": 0.0001,
-        "weight_decay": 1e-06,
-        "grad_clip_norm": 10.0,
+        "weight_decay": 1e-10,
+        "grad_clip_norm": 10,
         "betas": [
-            0.95,
-            0.999
+            0.9,
+            0.95
         ],
         "eps": 1e-08
     },
     "scheduler": {
-        "type": "diffuser",
-        "num_warmup_steps": 500,
-        "name": "cosine"
+        "type": "cosine_decay_with_warmup",
+        "num_warmup_steps": 1000,
+        "num_decay_steps": 30000,
+        "peak_lr": 0.0001,
+        "decay_lr": 2.5e-06
     },
     "eval": {
         "n_episodes": 50,