perf(inference): batched causal prefill attention + elementwise batching#189
Open
ohdearquant wants to merge 1 commit into
Open
perf(inference): batched causal prefill attention + elementwise batching#189ohdearquant wants to merge 1 commit into
ohdearquant wants to merge 1 commit into
Conversation
Replaces the per-token decode-attention loop in prefill with a single batched causal attention pass, and batches the elementwise prefill ops to eliminate per-token GPU dispatches. Removes an unnecessary unsafe in the attn fallback. Decode unchanged. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This was referenced Jun 4, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
batched causal prefill attention + elementwise batching
Stacked on #188. Replaces the per-token decode-attention loop in prefill with a
single batched causal-attention pass, and batches the elementwise prefill ops so they
no longer issue a GPU dispatch per token.
Why
After the chunked-prefill fix, the remaining per-token cost in prefill was the
attention loop and the elementwise ops running once per position. This collapses both
into batched passes over the prompt.
Result
Additional prefill speedup on top of PR1 (cumulative interleaved A/B tracked
internally; a fresh same-process A/B will be attached before merge rather than quoting
a number I can't reproduce in-session). Decode path unchanged; prefill argmax parity
preserved across runs.
Notes
unsafein the attention fallback.unwrap().