google-deepmind / google-deepmind/alphageometry
Inference does not use GPUs.
Open
- Dominant language
- Python
- Stars
- 4.9k
- Forks
- 572
- PR merge metrics
- No merged PRs in 30d
Description
I have updated Jax to the version support CUDA.
```
pip install --upgrade --user jax==0.4.6 jaxlib==0.4.6+cuda11.cudnn86 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
```
I also check that Jax now can detect GPU.
```
(alphageometry) [jpark3234@atl1-1-03-010-10-0 alphageometry]$ python -c "import jax; print(jax.devices())"
[StreamExecutorGpuDevice(id=0, process_index=0, slice_index=0)]
```
But when I run the code in run.sh, it does not utilize GPU (only few memcopy occurred in GPU).
Anyone has met/solved this problem?
Contributor guide
Assessment
This issue has not been assessed yet.