Repository navigation
Add Python 3.14 support and move to torch 2.14 with CUDA 13. - #1194
Merged
Merged
Conversation
Closed
vestoriabot
added a commit
to vestoriaai/mmjlab
that referenced
this pull request
Sep 30, 2026
`run_train` hard-coded `device="cpu"` whenever CUDA was not requested, leaving Apple Silicon users no path to GPU-accelerated policy training. Mirror the CUDA branch's auto-detection style: with no CUDA devices, pick `mps` when `torch.backends.mps.is_available()`, else `cpu`. MuJoCo physics stays on CPU in that case: the classic backend requires a CPU device and MuJoCo Warp has no Metal backend, so only policy inference/learning runs on MPS. On YAM lift-cube (vision, 64 envs) this cuts iteration time 20.4s -> 2.1s (~10x); pure-MLP tasks are ~2x slower on MPS than CPU, so CPU remains preferable there. Motivated by Apple Silicon training support; upstream mujocolab#1194's move to torch 2.14 keeps MPS a first-class torch target going forward. Co-Authored-By: Claude Code <noreply@anthropic.com>
Contributor
|
@kevinzakka is there a way to also add support for CUDA 12 alongside? Some server-side GPUs still maintain CUDA 12 and do not support CUDA 13 yet, and this'll create a problem for us to bump mjlab versions in the future. Do you see a possibility? |
Collaborator
Author
|
Hi @saikishor, can you open an issue so I can track it? |
Contributor
Sure thing. I'll create it as soon as I get access to my PC. Thank you |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds Python 3.14 support and moves to torch 2.14, which clears the remaining torch Dependabot alerts. PyTorch no longer publishes cu128 wheels past 2.11, so the cu128 extra is replaced with cu130; training now needs driver 580+.