Upload policy weights, train config and readme

README.md

@@ -1,22 +1,21 @@
 ---
-base_model: lerobot/smolvla_base
-datasets: Beilinghamburger/so100_vla_dataset_100
+datasets: Beilinghamburger/so100_vla_dataset
 library_name: lerobot
 license: apache-2.0
-model_name: smolvla
+model_name: pi0
 pipeline_tag: robotics
 tags:
-- lerobot
+- pi0
 - robotics
-- smolvla
+- lerobot
 ---
 
-# Model Card for smolvla
+# Model Card for pi0
 
 <!-- Provide a quick summary of what the model is/does. -->
 
 
-[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.
+[Pi0](https://huggingface.co/papers/2410.24164) is a generalist vision-language-action transformer that converts multimodal observations and text instructions into robot actions for zero-shot task transfer.
 
 
 This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).

config.json

@@ -1,6 +1,11 @@
 {
-    "type": "smolvla",
+    "type": "pi0",
     "n_obs_steps": 1,
+    "normalization_mapping": {
+        "VISUAL": "IDENTITY",
+        "STATE": "MEAN_STD",
+        "ACTION": "MEAN_STD"
+    },
     "input_features": {
         "observation.state": {
             "type": "STATE",
@@ -33,58 +38,40 @@
             ]
         }
     },
-    "device": "cuda",
+    "device": "cpu",
     "use_amp": false,
     "push_to_hub": true,
     "repo_id": "Beilinghamburger/smolvla_so100_vla",
     "private": null,
     "tags": null,
     "license": null,
-    "pretrained_path": null,
     "chunk_size": 50,
     "n_action_steps": 50,
-    "normalization_mapping": {
-        "VISUAL": "IDENTITY",
-        "STATE": "MEAN_STD",
-        "ACTION": "MEAN_STD"
-    },
     "max_state_dim": 32,
     "max_action_dim": 32,
     "resize_imgs_with_padding": [
-        512,
-        512
+        224,
+        224
     ],
     "empty_cameras": 0,
     "adapt_to_pi_aloha": false,
     "use_delta_joint_actions_aloha": false,
     "tokenizer_max_length": 48,
+    "proj_width": 1024,
     "num_steps": 10,
     "use_cache": true,
+    "attention_implementation": "eager",
     "freeze_vision_encoder": true,
-    "train_expert_only": true,
+    "train_expert_only": false,
     "train_state_proj": true,
-    "optimizer_lr": 0.0001,
+    "optimizer_lr": 2.5e-05,
     "optimizer_betas": [
         0.9,
         0.95
     ],
     "optimizer_eps": 1e-08,
     "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,
-    "rtc_config": null
+    "scheduler_decay_lr": 2.5e-06
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1660564e0274422231d942fe1dea8857151657ecbf80a572d101a39f0b632198
-size 1197789224
+oid sha256:27936bb0da4cc16781f1dc4a19419790a56c3f547707cd4c38ba78386954d94d
+size 7536025152

train_config.json

@@ -1,6 +1,6 @@
 {
     "dataset": {
-        "repo_id": "Beilinghamburger/so100_vla_dataset_100",
+        "repo_id": "Beilinghamburger/so100_vla_dataset",
         "root": null,
         "episodes": null,
         "image_transforms": {
@@ -57,20 +57,6 @@
                             1.5
                         ]
                     }
-                },
-                "affine": {
-                    "weight": 1.0,
-                    "type": "RandomAffine",
-                    "kwargs": {
-                        "degrees": [
-                            -5.0,
-                            5.0
-                        ],
-                        "translate": [
-                            0.05,
-                            0.05
-                        ]
-                    }
                 }
             }
         },
@@ -81,8 +67,13 @@
     },
     "env": null,
     "policy": {
-        "type": "smolvla",
+        "type": "pi0",
         "n_obs_steps": 1,
+        "normalization_mapping": {
+            "VISUAL": "IDENTITY",
+            "STATE": "MEAN_STD",
+            "ACTION": "MEAN_STD"
+        },
         "input_features": {
             "observation.state": {
                 "type": "STATE",
@@ -115,78 +106,60 @@
                 ]
             }
         },
-        "device": "cuda",
+        "device": "cpu",
         "use_amp": false,
         "push_to_hub": true,
         "repo_id": "Beilinghamburger/smolvla_so100_vla",
         "private": null,
         "tags": null,
         "license": null,
-        "pretrained_path": null,
         "chunk_size": 50,
         "n_action_steps": 50,
-        "normalization_mapping": {
-            "VISUAL": "IDENTITY",
-            "STATE": "MEAN_STD",
-            "ACTION": "MEAN_STD"
-        },
         "max_state_dim": 32,
         "max_action_dim": 32,
         "resize_imgs_with_padding": [
-            512,
-            512
+            224,
+            224
         ],
         "empty_cameras": 0,
         "adapt_to_pi_aloha": false,
         "use_delta_joint_actions_aloha": false,
         "tokenizer_max_length": 48,
+        "proj_width": 1024,
         "num_steps": 10,
         "use_cache": true,
+        "attention_implementation": "eager",
         "freeze_vision_encoder": true,
-        "train_expert_only": true,
+        "train_expert_only": false,
         "train_state_proj": true,
-        "optimizer_lr": 0.0001,
+        "optimizer_lr": 2.5e-05,
         "optimizer_betas": [
             0.9,
             0.95
         ],
         "optimizer_eps": 1e-08,
         "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,
-        "rtc_config": null
+        "scheduler_decay_lr": 2.5e-06
     },
-    "output_dir": "outputs/train/smolvla_so100_vla1215_new",
-    "job_name": "smolvla_so100_vla",
+    "output_dir": "outputs/train/pi_so100_vla1215",
+    "job_name": "pi_so100_vla",
     "resume": false,
     "seed": 1000,
     "num_workers": 4,
-    "batch_size": 32,
-    "steps": 40000,
-    "eval_freq": 10000,
+    "batch_size": 4,
+    "steps": 1,
+    "eval_freq": 100,
     "log_freq": 200,
     "save_checkpoint": true,
-    "save_freq": 10000,
+    "save_freq": 100,
     "use_policy_training_preset": true,
     "optimizer": {
         "type": "adamw",
-        "lr": 0.0001,
+        "lr": 2.5e-05,
         "weight_decay": 1e-10,
-        "grad_clip_norm": 10,
+        "grad_clip_norm": 10.0,
         "betas": [
             0.9,
             0.95
@@ -197,7 +170,7 @@
         "type": "cosine_decay_with_warmup",
         "num_warmup_steps": 1000,
         "num_decay_steps": 30000,
-        "peak_lr": 0.0001,
+        "peak_lr": 2.5e-05,
         "decay_lr": 2.5e-06
     },
     "eval": {
@@ -213,7 +186,5 @@
         "notes": null,
         "run_id": null,
         "mode": null
-    },
-    "checkpoint_path": null,
-    "rename_map": {}
+    }
 }