Upload policy weights, train config and readme

README.md

@@ -6,8 +6,8 @@ license: apache-2.0
 model_name: smolvla
 pipeline_tag: robotics
 tags:
-- smolvla
 - lerobot
+- smolvla
 - robotics
 ---
 

config.json

@@ -40,7 +40,7 @@
     "private": null,
     "tags": null,
     "license": null,
-    "pretrained_path": "./lerobot_smolvla_base_migrated",
+    "pretrained_path": null,
     "chunk_size": 50,
     "n_action_steps": 50,
     "normalization_mapping": {
@@ -70,17 +70,17 @@
     ],
     "optimizer_eps": 1e-08,
     "optimizer_weight_decay": 1e-10,
-    "optimizer_grad_clip_norm": 10.0,
+    "optimizer_grad_clip_norm": 10,
     "scheduler_warmup_steps": 1000,
     "scheduler_decay_steps": 30000,
-    "scheduler_decay_lr": 2.5e-06,
+    "scheduler_decay_lr": 0.025,
     "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
-    "load_vlm_weights": true,
+    "load_vlm_weights": false,
     "add_image_special_tokens": false,
     "attention_mode": "cross_attn",
-    "prefix_length": 0,
-    "pad_language_to": "max_length",
-    "num_expert_layers": 0,
+    "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,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea3eab3e960358290831fe5c23193dd5a94d6ef907b50d443c83b1d58f5864bf
-size 906712520
+oid sha256:a99f156bd0a564124b32f77323bdb8b9bbcd1691f8e5db9481f70aef768238fb
+size 1197789224

train_config.json

@@ -1,7 +1,7 @@
 {
     "dataset": {
         "repo_id": "Calvert0921/SmolVLA_LiftBlackCube5_Franka_100",
-        "root": null,
+        "root": "./dataset/Calvert0921/SmolVLA_LiftBlackCube5_Franka_100",
         "episodes": null,
         "image_transforms": {
             "enable": false,
@@ -108,7 +108,7 @@
         "private": null,
         "tags": null,
         "license": null,
-        "pretrained_path": "./lerobot_smolvla_base_migrated",
+        "pretrained_path": null,
         "chunk_size": 50,
         "n_action_steps": 50,
         "normalization_mapping": {
@@ -138,17 +138,17 @@
         ],
         "optimizer_eps": 1e-08,
         "optimizer_weight_decay": 1e-10,
-        "optimizer_grad_clip_norm": 10.0,
+        "optimizer_grad_clip_norm": 10,
         "scheduler_warmup_steps": 1000,
         "scheduler_decay_steps": 30000,
-        "scheduler_decay_lr": 2.5e-06,
+        "scheduler_decay_lr": 0.025,
         "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
-        "load_vlm_weights": true,
+        "load_vlm_weights": false,
         "add_image_special_tokens": false,
         "attention_mode": "cross_attn",
-        "prefix_length": 0,
-        "pad_language_to": "max_length",
-        "num_expert_layers": 0,
+        "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,
@@ -161,9 +161,9 @@
     "seed": 1000,
     "num_workers": 4,
     "batch_size": 1,
-    "steps": 1000,
+    "steps": 10000000,
     "eval_freq": 20000,
-    "log_freq": 200,
+    "log_freq": 10,
     "save_checkpoint": true,
     "save_freq": 20000,
     "use_policy_training_preset": true,
@@ -171,7 +171,7 @@
         "type": "adamw",
         "lr": 0.0001,
         "weight_decay": 1e-10,
-        "grad_clip_norm": 10.0,
+        "grad_clip_norm": 10,
         "betas": [
             0.9,
             0.95
@@ -183,7 +183,7 @@
         "num_warmup_steps": 1000,
         "num_decay_steps": 30000,
         "peak_lr": 0.0001,
-        "decay_lr": 2.5e-06
+        "decay_lr": 0.025
     },
     "eval": {
         "n_episodes": 50,