Mac different results

by l0d0v1c - opened Jun 6

Jun 6

I tried with mac M2 (using https://download.pytorch.org/whl/nightly/cpu becaus bfloat16 is not supported on MPS). It works but I get different results:

tensor([[0.5709, 0.1221],
[0.0952, 0.3745]])

insead of

tensor([[0.6175, 0.2535],
[0.2298, 0.5792]])

in your example.
Anyway very interesting work! Thank you.

vaibhavad

McGill NLP Group org Jul 1

Hi @l0d0v1c ,

Apologies for the late response. Unfortunately, we haven't extensively tested our models beyond NVIDIA GPUs using BF16, so the behaviour might be unpredictable. However, using float32 should give you similar results on M2 and NVIDIA.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment