Upload policy weights, train config and readme

README.md

@@ -1,14 +1,14 @@
 ---
 base_model: lerobot/smolvla_base
-datasets: RonPlusSign/RLBench-PutRubbishInBin-joint_positions
+datasets: RonPlusSign/RLBench-LeRobot-v3-PutRubbishInBin
 library_name: lerobot
 license: apache-2.0
 model_name: smolvla
 pipeline_tag: robotics
 tags:
-- lerobot
-- smolvla
 - robotics
+- smolvla
+- lerobot
 ---
 
 # Model Card for smolvla

config.json

@@ -53,7 +53,7 @@
         "action": {
             "type": "ACTION",
             "shape": [
-                8
+                7
             ]
         }
     },
@@ -64,9 +64,9 @@
     "private": null,
     "tags": null,
     "license": null,
-    "pretrained_path": "RonPlusSign/smolvla_base",
+    "pretrained_path": "outputs/train/smolvla_finetuned_rubbish/checkpoints/last/pretrained_model",
     "chunk_size": 50,
-    "n_action_steps": 50,
+    "n_action_steps": 1,
     "normalization_mapping": {
         "VISUAL": "IDENTITY",
         "STATE": "MEAN_STD",
@@ -98,16 +98,16 @@
     "scheduler_warmup_steps": 1000,
     "scheduler_decay_steps": 30000,
     "scheduler_decay_lr": 2.5e-06,
-    "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
+    "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Instruct",
     "load_vlm_weights": true,
     "add_image_special_tokens": false,
     "attention_mode": "cross_attn",
     "prefix_length": 0,
-    "pad_language_to": "max_length",
-    "num_expert_layers": 0,
-    "num_vlm_layers": 16,
+    "pad_language_to": "longest",
+    "num_expert_layers": -1,
+    "num_vlm_layers": 0,
     "self_attn_every_n_layers": 2,
-    "expert_width_multiplier": 0.75,
+    "expert_width_multiplier": 0.5,
     "min_period": 0.004,
     "max_period": 4.0
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3fcf2cfe93e6ee77831b0e6c8cb9a757be6caaefa67387db07f91ff62294f60
-size 906712520
+oid sha256:1106fb447275318a9e1f3b013981a08414bd89f5ab55665c09d6cac79b9b5c62
+size 1218047032

train_config.json

@@ -1,6 +1,6 @@
 {
     "dataset": {
-        "repo_id": "RonPlusSign/RLBench-PutRubbishInBin-joint_positions",
+        "repo_id": "RonPlusSign/RLBench-LeRobot-v3-PutRubbishInBin",
         "root": null,
         "episodes": null,
         "image_transforms": {
@@ -121,7 +121,7 @@
             "action": {
                 "type": "ACTION",
                 "shape": [
-                    8
+                    7
                 ]
             }
         },
@@ -132,9 +132,9 @@
         "private": null,
         "tags": null,
         "license": null,
-        "pretrained_path": "RonPlusSign/smolvla_base",
+        "pretrained_path": "outputs/train/smolvla_finetuned_rubbish/checkpoints/last/pretrained_model",
         "chunk_size": 50,
-        "n_action_steps": 50,
+        "n_action_steps": 1,
         "normalization_mapping": {
             "VISUAL": "IDENTITY",
             "STATE": "MEAN_STD",
@@ -166,30 +166,30 @@
         "scheduler_warmup_steps": 1000,
         "scheduler_decay_steps": 30000,
         "scheduler_decay_lr": 2.5e-06,
-        "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
+        "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Instruct",
         "load_vlm_weights": true,
         "add_image_special_tokens": false,
         "attention_mode": "cross_attn",
         "prefix_length": 0,
-        "pad_language_to": "max_length",
-        "num_expert_layers": 0,
-        "num_vlm_layers": 16,
+        "pad_language_to": "longest",
+        "num_expert_layers": -1,
+        "num_vlm_layers": 0,
         "self_attn_every_n_layers": 2,
-        "expert_width_multiplier": 0.75,
+        "expert_width_multiplier": 0.5,
         "min_period": 0.004,
         "max_period": 4.0
     },
-    "output_dir": "outputs/train/smolvla_finetuned_rubbish_joints",
+    "output_dir": "outputs/train/smolvla_finetuned_rubbish",
     "job_name": "smolvla",
-    "resume": false,
+    "resume": true,
     "seed": 1000,
     "num_workers": 4,
-    "batch_size": 64,
-    "steps": 20000,
+    "batch_size": 32,
+    "steps": 60000,
     "eval_freq": 20000,
     "log_freq": 200,
     "save_checkpoint": true,
-    "save_freq": 20000,
+    "save_freq": 10000,
     "use_policy_training_preset": true,
     "optimizer": {
         "type": "adamw",
@@ -220,7 +220,7 @@
         "project": "lerobot",
         "entity": null,
         "notes": null,
-        "run_id": "o3rqxmof",
+        "run_id": "gwlgotby",
         "mode": null
     }
 }