Skip to content

Conversation

@abcdabcd987
Copy link
Member

In my development setup, I have different machines with different GPUs. They share the home directory on a network filesystem. When I switch between machines, since the JIT compilation flags change, I'll have to recompile kernels every time.

One solution is to specify the same TORCH_CUDA_ARCH_LIST every time. However, I keep forgetting that.

Another solution, as proposed in this PR, is to put different arch list in different cache directory.

@abcdabcd987 abcdabcd987 requested a review from yzh119 October 29, 2024 19:19
@yzh119 yzh119 merged commit cdc12c3 into flashinfer-ai:main Oct 29, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants