Replies: 1 comment 1 reply
-
|
This was committed. |
Beta Was this translation helpful? Give feedback.
1 reply
-
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
The change allows simplifying away lowering conversions (e.g.
f32->bf16->f32), and was seen to lead to large increases in performance on some models (~10%).This can cause casts being ignored, and overall precision higher than expected, but it's already the case for XLA compiler (e.g. elementwise ops and matmuls can be already performed in higher precision than specified in HLO)
Beta Was this translation helpful? Give feedback.
All reactions