Skip to content

Allow f16/bf16 in custom SDPA#237

Open
GregoryComer wants to merge 1 commit into
huggingface:mainfrom
GregoryComer:sdpa-half
Open

Allow f16/bf16 in custom SDPA#237
GregoryComer wants to merge 1 commit into
huggingface:mainfrom
GregoryComer:sdpa-half

Conversation

@GregoryComer

Copy link
Copy Markdown

The current module swapped SDPA always runs SDPA in f32, up-converting activations if necessary. We now have native support for half-precision SDPA in ET, so relax the assertions and remove the casts.

@GregoryComer
GregoryComer marked this pull request as ready for review July 9, 2026 00:02
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