-
Notifications
You must be signed in to change notification settings - Fork 4.1k
Open
Labels
bugSomething isn’t working correctlySomething isn’t working correctlydatasetIssues regarding data inputs, processing, or datasetsIssues regarding data inputs, processing, or datasetsdependenciesConcerns about external packages, libraries, or versioningConcerns about external packages, libraries, or versioningdocumentationImprovements or fixes to the project’s docsImprovements or fixes to the project’s docsvisualizationIssues about visual output, graphs, or data visualizationIssues about visual output, graphs, or data visualization
Description
Ticket Type
🐛 Bug Report
Environment & System Info
- LeRobot Version: 0.5.1 (autoinstall)
- OS: Ubuntu 24.04 LTS
- Python: 3.12 (Miniconda)
- PyTorch Version: 2.10.0+cu128
- GPU: NVIDIA GeForce RTX 5070 Laptop (Driver 580.126.09)
Description
Simply uploading a lerobot-record dataset using default methods causes the dataset to be unviewable by the dataset visualizer.
- The visualizer UI fails with:
"Something went wrong. Failed to fetch dataset info: 401"(or Error 404). - The CLI utility
lerobot-dataset-vizcrashes with aRuntimeErrorregardinglibtorchcodec.
It appears to be a library mismatch between the environment's FFmpeg/PyTorch version and what torchcodec expects.
Context & Reproduction
The crash occurs immediately after "Logging to Rerun" when the dataloader begins fetching frames.
Command:
lerobot-dataset-viz --repo-id thesimbotbro/eval_PickAndPlace_30Hz_Seth3 --episode-index 0Relevant Logs & Stack Trace
INFO 2026-03-21 23:10:12 aset_viz.py:115 Starting Rerun
[2026-03-22T03:10:12Z INFO re_grpc_server] Listening for gRPC connections on 0.0.0.0:9876.
...
Traceback (most recent call last):
File ".../lerobot/datasets/video_utils.py", line 152, in decode_video_frames
return decode_video_frames_torchcodec(video_path, timestamps, tolerance_s)
...
File ".../torchcodec/_core/ops.py", line 69, in load_torchcodec_shared_libraries
raise RuntimeError(
RuntimeError: Could not load libtorchcodec. Likely causes:
1. FFmpeg is not properly installed in your environment. We support versions 4, 5, 6 and 7.
2. The PyTorch version (2.10.0+cu128) is not compatible with this version of TorchCodec.
3. Another runtime dependency.
[start of libtorchcodec loading traceback]
FFmpeg version 7: Could not load this library: .../libtorchcodec_core7.so
FFmpeg version 6: Could not load this library: .../libtorchcodec_core6.so
FFmpeg version 5: Could not load this library: .../libtorchcodec_core5.so
FFmpeg version 4: Could not load this library: .../libtorchcodec_core4.so
[end of libtorchcodec loading traceback]
Checklist
- I have searched existing tickets to ensure this isn't a duplicate.
- I am using the latest version of the
mainbranch. - I have verified this is not an environment-specific problem.
Additional Info / Workarounds
It seems that torchcodec is unable to find the shared objects for FFmpeg versions 4-7 in the current environment path, or there is a version mismatch with PyTorch 2.10.0+cu128.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
bugSomething isn’t working correctlySomething isn’t working correctlydatasetIssues regarding data inputs, processing, or datasetsIssues regarding data inputs, processing, or datasetsdependenciesConcerns about external packages, libraries, or versioningConcerns about external packages, libraries, or versioningdocumentationImprovements or fixes to the project’s docsImprovements or fixes to the project’s docsvisualizationIssues about visual output, graphs, or data visualizationIssues about visual output, graphs, or data visualization