RuntimeError: Could not load libtorchcodec during lerobot/scripts/train.py script

[shrutichakraborty]

System Info

- `lerobot` version: 0.1.0
- Platform: Linux-6.8.0-57-generic-x86_64-with-glibc2.35
- Python version: 3.10.13
- Huggingface_hub version: 0.29.3
- Dataset version: 3.4.1
- Numpy version: 1.26.4
- PyTorch version (GPU?): 2.5.1+cu124 (True)
- Cuda version: 12040


Additionally: 

ffmpeg version : 7.1.1
TorchCodec version : 0.2.1

Information

• One of the scripts in the examples/ folder of LeRobot
• My own task or dataset (give details below)

Reproduction

Install leRobot from the main documentation as follows :

conda create -n lerobot python=3.10 -y
conda activate lerobot
git clone https://github.com/huggingface/lerobot.git ~/lerobot
pip install --no-binary=av -e
pip install torchvision==0.20.1
conda install -c conda-forge 'ffmpeg>=7.0' -y

After collecting a dataset, run lerobot/scripts/train.py script

Expected behavior

Hello all!

I am getting started with the lerobot so100 arm and have had a few issues.

The first was the same as the issue in #883 in running the control_robot.py script which I solved (or bypassed) by following remi cadene's response to do pip install torchvision==0.20.1 and also conda install -c conda-forge 'ffmpeg>=7.0' -y after doing pip install --no-binary=av -e . This allowed me to successfully run the control_robot.py script successfully. However, then I tried to collect a dataset and run a training with the lerobot/scripts/train.py script and I encountered the following issue :

from torchcodec.decoders._core.video_decoder_ops import (
  File "/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/decoders/_core/video_decoder_ops.py", line 59, in <module>
    load_torchcodec_extension()
  File "/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/decoders/_core/video_decoder_ops.py", line 44, in load_torchcodec_extension
    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 (2.5.1+cu124) is not compatible with
             this version of TorchCodec. Refer to the version compatibility
             table:
             https://github.com/pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec.
          3. Another runtime dependency; see exceptions below.
        The following exceptions were raised as we tried to load libtorchcodec:
        
[start of libtorchcodec loading traceback]
/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/libtorchcodec7.so: undefined symbol: _ZNK3c1011StorageImpl27throw_data_ptr_access_errorEv
libavutil.so.58: cannot open shared object file: No such file or directory
libavutil.so.57: cannot open shared object file: No such file or directory
/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/libtorchcodec4.so: undefined symbol: _ZNK3c1011StorageImpl27throw_data_ptr_access_errorEv
[end of libtorchcodec loading traceback].

It seems that I have some issues with the torchcodecand ffmpeg versions not being compatible. Checking their versions gives me:

ffmpeg version 7.1.1 Copyright (c) 2000-2025 the FFmpeg developers
built with gcc 13.3.0 (conda-forge gcc 13.3.0-2)
configuration: --prefix=/home/moonshot/miniconda3/envs/lerobot --cc=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-cc --cxx=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-c++ --nm=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-nm --ar=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-ar --disable-doc --enable-openssl --enable-demuxer=dash --enable-hardcoded-tables --enable-libfreetype --enable-libharfbuzz --enable-libfontconfig --enable-libopenh264 --enable-libdav1d --disable-gnutls --enable-libmp3lame --enable-libvpx --enable-libass --enable-pthreads --enable-alsa --enable-libpulse --enable-vaapi --enable-libopenvino --enable-gpl --enable-libx264 --enable-libx265 --enable-libaom --enable-libsvtav1 --enable-libxml2 --enable-pic --enable-shared --disable-static --enable-version3 --enable-zlib --enable-libvorbis --enable-libopus --enable-librsvg --enable-ffplay --pkg-config=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/pkg-config
libavutil      59. 39.100 / 59. 39.100
libavcodec     61. 19.101 / 61. 19.101
libavformat    61.  7.100 / 61.  7.100
libavdevice    61.  3.100 / 61.  3.100
libavfilter    10.  4.100 / 10.  4.100
libswscale      8.  3.100 /  8.  3.100
libswresample   5.  3.100 /  5.  3.100
libpostproc    58.  3.100 / 58.  3.100

And TorchCodec version 0.2.1.

Could anyone suggest the right versions to install and whether I should downgrade ffmpeg ?

[imstevenpmwork]

Try with latest main and new installation instructions (:

[gpthimble]

I have the exact same issue. But I think it is due to some problem with the torchcodec.
When I debug this issue , I tried to start with a complete new environment, install torch and then ffmpeg and finally the torchcodec. And I didn't install the lerobot.
When I tried to run "import torchcodec" in python interpreter, the same error still occured , I have tried different ffmpeg version, no fix yet.

[CarolinePascal]

Hi @gpthimble !

Can you confirm that you installed ffmpeg using conda during your second attempt ? This is actually a requirement to make torchcodec work correctly.
If so, could you provide us with the output of the following commands (which have to be run in your LeRobot conda environment) :

which ffmpeg
ffmpeg -version
ffmpeg -encoders | grep libsvtav1

Thanks,

Caroline.

[gpthimble]

Hi @gpthimble !

Can you confirm that you installed ffmpeg using conda during your second attempt ? This is actually a requirement to make torchcodec work correctly. If so, could you provide us with the output of the following commands (which have to be run in your LeRobot conda environment) :

which ffmpeg
ffmpeg -version
ffmpeg -encoders | grep libsvtav1

Thanks,

Caroline.

I have tried the new instructions in the readme, build a new enviroment, install the ffmpeg 7.1.1.
Everything works now! Thank you!

[gpthimble]

System Info

• lerobot version: 0.1.0
• Platform: Linux-6.8.0-57-generic-x86_64-with-glibc2.35
• Python version: 3.10.13
• Huggingface_hub version: 0.29.3
• Dataset version: 3.4.1
• Numpy version: 1.26.4
• PyTorch version (GPU?): 2.5.1+cu124 (True)
• Cuda version: 12040

Additionally:

ffmpeg version : 7.1.1
TorchCodec version : 0.2.1

Information

• One of the scripts in the examples/ folder of LeRobot[ ] My own task or dataset (give details below)

Reproduction

Install leRobot from the main documentation as follows :

conda create -n lerobot python=3.10 -y conda activate lerobot git clone https://github.com/huggingface/lerobot.git ~/lerobot pip install --no-binary=av -e pip install torchvision==0.20.1 conda install -c conda-forge 'ffmpeg>=7.0' -y

After collecting a dataset, run lerobot/scripts/train.py script

Expected behavior

Hello all!

I am getting started with the lerobot so100 arm and have had a few issues.

The first was the same as the issue in #883 in running the control_robot.py script which I solved (or bypassed) by following remi cadene's response to do pip install torchvision==0.20.1 and also conda install -c conda-forge 'ffmpeg>=7.0' -y after doing pip install --no-binary=av -e . This allowed me to successfully run the control_robot.py script successfully. However, then I tried to collect a dataset and run a training with the lerobot/scripts/train.py script and I encountered the following issue :

from torchcodec.decoders._core.video_decoder_ops import (
  File "/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/decoders/_core/video_decoder_ops.py", line 59, in <module>
    load_torchcodec_extension()
  File "/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/decoders/_core/video_decoder_ops.py", line 44, in load_torchcodec_extension
    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 (2.5.1+cu124) is not compatible with
             this version of TorchCodec. Refer to the version compatibility
             table:
             https://github.com/pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec.
          3. Another runtime dependency; see exceptions below.
        The following exceptions were raised as we tried to load libtorchcodec:
        
[start of libtorchcodec loading traceback]
/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/libtorchcodec7.so: undefined symbol: _ZNK3c1011StorageImpl27throw_data_ptr_access_errorEv
libavutil.so.58: cannot open shared object file: No such file or directory
libavutil.so.57: cannot open shared object file: No such file or directory
/home/moonshot/miniconda3/envs/lerobot/lib/python3.10/site-packages/torchcodec/libtorchcodec4.so: undefined symbol: _ZNK3c1011StorageImpl27throw_data_ptr_access_errorEv
[end of libtorchcodec loading traceback].

It seems that I have some issues with the torchcodecand ffmpeg versions not being compatible. Checking their versions gives me:

ffmpeg version 7.1.1 Copyright (c) 2000-2025 the FFmpeg developers
built with gcc 13.3.0 (conda-forge gcc 13.3.0-2)
configuration: --prefix=/home/moonshot/miniconda3/envs/lerobot --cc=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-cc --cxx=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-c++ --nm=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-nm --ar=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/x86_64-conda-linux-gnu-ar --disable-doc --enable-openssl --enable-demuxer=dash --enable-hardcoded-tables --enable-libfreetype --enable-libharfbuzz --enable-libfontconfig --enable-libopenh264 --enable-libdav1d --disable-gnutls --enable-libmp3lame --enable-libvpx --enable-libass --enable-pthreads --enable-alsa --enable-libpulse --enable-vaapi --enable-libopenvino --enable-gpl --enable-libx264 --enable-libx265 --enable-libaom --enable-libsvtav1 --enable-libxml2 --enable-pic --enable-shared --disable-static --enable-version3 --enable-zlib --enable-libvorbis --enable-libopus --enable-librsvg --enable-ffplay --pkg-config=/home/conda/feedstock_root/build_artifacts/ffmpeg_1741820412024/_build_env/bin/pkg-config
libavutil      59. 39.100 / 59. 39.100
libavcodec     61. 19.101 / 61. 19.101
libavformat    61.  7.100 / 61.  7.100
libavdevice    61.  3.100 / 61.  3.100
libavfilter    10.  4.100 / 10.  4.100
libswscale      8.  3.100 /  8.  3.100
libswresample   5.  3.100 /  5.  3.100
libpostproc    58.  3.100 / 58.  3.100

And TorchCodec version 0.2.1.

Could anyone suggest the right versions to install and whether I should downgrade ffmpeg ?

I think you should try different version of Torchcodec. version 0.2.1 is for the torch version 2.6 not 2.5.
Check this link: https://github.com/pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec

[CarolinePascal]

Thanks for pointing that out !

We're going to solve all ffmpeg and torchcodec related versioning issues very soon, and I think we'll update our pyproject.toml by then ;)

I'm closing the issue as it is solved now.

Best,

Caroline.

[gpthimble]

Thanks for pointing that out !

We're going to solve all ffmpeg and torchcodec related versioning issues very soon, and I think we'll update our pyproject.toml by then ;)

I'm closing the issue as it is solved now.

Best,

Caroline.

I have also tried lerobot project on an AMD GPU with rocm, with no luck. I think the torch itself should work without problem, but it seems that the package torchcodec is not compactable with rocm version of torch. I will post an issue on the torchcodec repository about that. Really hope this can work on an amd gpu!