Skip to content

Conversation

catwell
Copy link
Member

@catwell catwell commented Oct 9, 2024

  • we check only the test device, not the machine in general (e.g. case where there are two different GPUs on a machine)
  • we don't want emulated bfloat16 (e.g. CPU)

(PyTorch for reference: https://github.com/pytorch/pytorch/blob/de4c2a3b4e89d96334dc678d1c3f2ae51a6630a0/torch/cuda/__init__.py#L132-L168)

- we check only the test device, not the machine in general
- we don't want emulated bfloat16 (e.g. CPU)
@catwell catwell force-pushed the pr/fix-bf16-detection branch from eef07ec to 4a1956a Compare October 9, 2024 08:51
@catwell catwell requested a review from deltheil October 9, 2024 08:54
@catwell catwell merged commit 2796117 into main Oct 9, 2024
3 checks passed
@catwell catwell deleted the pr/fix-bf16-detection branch October 9, 2024 09:20
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