Add perf model for vllm::rocm_aiter_sparse_attn_indexer (GLM5) - #937
Open
spandoesai wants to merge 2 commits into
Open
spandoesai wants to merge 2 commits into
spandoesai wants to merge 2 commits into
Conversation
The DeepSeek sparse-attention "lightning indexer" kernel previously had no perf model, so it reported 0 FLOPs / 0 bytes and was excluded from roofline scoring despite being a sizeable fraction of decode time. Add rocm_aiter_sparse_attn_indexer (subclassing InferenceAttention to reuse the vLLM iteration-annotation parsing for exact QK-pair aggregates). FLOPs count the fp8 MQA index-score QK only (no value aggregation). Bytes reflect the fused MQA traffic: keys read once across the H index heads, split into cached fp8 rows (key + per-block scales) and the current-chunk bf16 key, plus the fp32 weights read and the top-k index write. Register the op in the pseudo-op mapping, and add it to the inference report's annotation name-filter so the kernel receives the seqlen aggregates its perf model needs. Co-authored-by: Cursor <cursoragent@cursor.com>
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The DeepSeek sparse-attention "lightning indexer" kernel had no perf model, so it reported 0 FLOPs/bytes and was dropped from roofline scoring despite being a meaningful chunk of GLM5 decode time. This adds rocm_aiter_sparse_attn_indexer (subclassing InferenceAttention to reuse the vLLM annotation parsing for exact QK-pair aggregates).