Upload dryrun/replay_episode.log with huggingface_hub
Browse files- dryrun/replay_episode.log +28 -0
dryrun/replay_episode.log
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[2026-08-30 22:10:30,564] INFO replay_lerobot_episode: episode 0: 414 rows
|
| 2 |
+
[2026-08-30 22:10:30,564] INFO replay_lerobot_episode: task: grab the bottle
|
| 3 |
+
[2026-08-30 22:10:30,663] INFO replay_lerobot_episode: video frames: 414
|
| 4 |
+
[2026-08-30 22:10:30,663] INFO root: Waiting for server at ws://127.0.0.1:5694...
|
| 5 |
+
[2026-08-30 22:10:30,667] WARNING root:
|
| 6 |
+
============================================================
|
| 7 |
+
[TRAIN/TEST CONSISTENCY CHECK] read before trusting results
|
| 8 |
+
------------------------------------------------------------
|
| 9 |
+
Make sure the EVAL observation matches TRAINING for:
|
| 10 |
+
- state : whether proprioceptive state is fed (use_state)
|
| 11 |
+
and its dimension / ordering
|
| 12 |
+
- img size : resize / crop resolution (e.g. 224x224)
|
| 13 |
+
- img count: how many camera views are fed to the model
|
| 14 |
+
- img order: the ordering of those camera views
|
| 15 |
+
- horizon : action chunk size / action horizon
|
| 16 |
+
A mismatch on ANY of these silently lowers the success rate.
|
| 17 |
+
Cross-check the values below against your training config.
|
| 18 |
+
The client will NOT infer or reorder camera views for you.
|
| 19 |
+
============================================================
|
| 20 |
+
[2026-08-30 22:10:30,667] WARNING root: [TRAIN/TEST CONSISTENCY CHECK] server metadata: {'env': 'starvla_policy_server', 'ckpt_path': '/225010261/StarVLA/src/starVLA/results/Checkpoints/starvla_qwenoft_g1_sonic_smoke/checkpoints/steps_1000_pytorch_model.pt', 'action_chunk_size': 8, 'available_unnorm_keys': ['new_embodiment'], 'default_unnorm_key': 'new_embodiment', 'training_data_mix': 'unitree_g1_test_sonic', 'training_obs_image_size': [224, 224], 'eval_image_contract': 'Eval clients must explicitly choose image count and order. The server does not infer or reorder camera views from training config.', 'action_keys': ['action.motion_token', 'action.left_hand_joints', 'action.right_hand_joints'], 'state_keys': ['state.left_leg', 'state.right_leg', 'state.waist', 'state.left_arm', 'state.left_hand', 'state.right_arm', 'state.right_hand', 'state.left_wrist_pos', 'state.left_wrist_abs_quat', 'state.right_wrist_pos', 'state.right_wrist_abs_quat', 'state.root_orientation', 'state.projected_gravity', 'state.cpp_rotation_offset', 'state.init_base_quat']}
|
| 21 |
+
[2026-08-30 22:10:30,667] INFO replay_lerobot_episode: connected; unnorm_key=new_embodiment
|
| 22 |
+
[2026-08-30 22:10:30,827] INFO root: [TRAIN/TEST CONSISTENCY CHECK] eval request image_count=1 image_shapes=[(224, 224)] server_metadata={'env': 'starvla_policy_server', 'ckpt_path': '/225010261/StarVLA/src/starVLA/results/Checkpoints/starvla_qwenoft_g1_sonic_smoke/checkpoints/steps_1000_pytorch_model.pt', 'action_chunk_size': 8, 'available_unnorm_keys': ['new_embodiment'], 'default_unnorm_key': 'new_embodiment', 'training_data_mix': 'unitree_g1_test_sonic', 'training_obs_image_size': [224, 224], 'eval_image_contract': 'Eval clients must explicitly choose image count and order. The server does not infer or reorder camera views from training config.', 'action_keys': ['action.motion_token', 'action.left_hand_joints', 'action.right_hand_joints'], 'state_keys': ['state.left_leg', 'state.right_leg', 'state.waist', 'state.left_arm', 'state.left_hand', 'state.right_arm', 'state.right_hand', 'state.left_wrist_pos', 'state.left_wrist_abs_quat', 'state.right_wrist_pos', 'state.right_wrist_abs_quat', 'state.root_orientation', 'state.projected_gravity', 'state.cpp_rotation_offset', 'state.init_base_quat']}. The benchmark interface is responsible for camera order; verify this order manually against the checkpoint training setup.
|
| 23 |
+
[2026-08-30 22:10:49,572] INFO replay_lerobot_episode: ...10/30 frames (avg 1852.6 ms)
|
| 24 |
+
[2026-08-30 22:11:09,266] INFO replay_lerobot_episode: ...20/30 frames (avg 1871.1 ms)
|
| 25 |
+
[2026-08-30 22:11:28,765] INFO replay_lerobot_episode: ...30/30 frames (avg 1865.5 ms)
|
| 26 |
+
[2026-08-30 22:11:28,766] INFO replay_lerobot_episode: replay done: 30 inferences, avg latency 1865.5 ms
|
| 27 |
+
[2026-08-30 22:11:28,766] INFO replay_lerobot_episode: pred-vs-recorded MSE: motion_token=0.0126 left_hand=0.0000 right_hand=0.0000
|
| 28 |
+
[2026-08-30 22:11:28,768] INFO replay_lerobot_episode: OK ✅ replay produced 64/7/7 splits for 30 real frames
|