CUDA version mismatch

[lukemovement]

The documentation has recently been updated to instruct the installation of CUDA 12.3, however @tensorflow/tfjs-node-gpu seems to be looking for version 11.

2024-08-24 22:24:04.039287: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda:/usr/local/cuda/lib64:

I managed to resolve the first missing file using the command below

sudo ln -s /usr/local/cuda/targets/x86_64-linux/lib/libcudart.so.12 /usr/local/cuda/targets/x86_64-linux/lib/libcudart.so.11.0

For TFJS should we still be using CUDA 11?

[shmishra99]

Hi @lukemovement ,

We appreciate you bringing this issue to our attention. Currently tfjs-node-gpu 1.2.4 or later supports NVIDIA® GPU drivers | >450.x, CUDA® Toolkit | 11.2 and cuDNN SDK | 8.1.0. Please refer to the official documentation for tfjs-node-gpu for detailed instructions.

Thank You!!

[lukemovement]

Hi @lukemovement ,

We appreciate you bringing this issue to our attention. Currently tfjs-node-gpu 1.2.4 or later supports NVIDIA® GPU drivers | >450.x, CUDA® Toolkit | 11.2 and cuDNN SDK | 8.1.0. Please refer to the official documentation for tfjs-node-gpu for detailed instructions.

Thank You!!

Managed to install it on 22.04 follow this guide. Might be worth adding it to the documentation considering 20.04 is on its way out?

[shmishra99]

I’m pleased to hear that your issue has been resolved. Thank you for your suggestion. As a reminder, our documentation specifies that for tfjs/tfjs-node installations with GPU drivers, Ubuntu 14.04 or later versions are currently supported. To install the latest TensorFlow version, we recommend Ubuntu 16.04 or newer. Please let us know if you have any further questions.

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No