NPUW: Support batched prefill scoring via row-by-row unrolling#36375
Draft
dylanneve1 wants to merge 1 commit into
Draft
NPUW: Support batched prefill scoring via row-by-row unrolling#36375dylanneve1 wants to merge 1 commit into
dylanneve1 wants to merge 1 commit into
Conversation
The NPUW LLM pipeline compiles its prefill/generate models with a static batch size of 1, so a batched [N, seq] input (e.g. GenAI's TextRerankPipeline scoring N documents in one infer) fails with a shape mismatch in pad_position_ids(). Accept batch > 1 at the LLMInferRequest boundary and unroll it: each row is sliced (zero-copy ROI view) and scored in its own prefill pass over the unchanged batch-1 compiled models, with the KV-cache state reset between rows; per-row logits are aggregated into one [N, ...] output tensor. Results are identical to per-row inference since batch rows are independent. Batched generation stays unsupported and is rejected with a clear error.
github-actions
Bot
added
category: NPU
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Details:
The NPUW LLM pipeline compiles its prefill/generate models with a static batch size of 1, so a batched [N, seq] input (e.g. GenAI's TextRerankPipeline scoring N documents in one infer) fails with a shape mismatch in pad_position_ids().
Accept batch > 1 at the LLMInferRequest boundary and unroll it: each row is sliced (zero-copy ROI view) and scored in its own prefill pass over the unchanged batch-1 compiled models, with the KV-cache state reset between rows; per-row logits are aggregated into one [N, ...] output tensor. Results are identical to per-row inference since batch rows are independent. Batched generation stays unsupported and is rejected with a clear error.
Tickets:
AI Assistance: