prefill: qwen35moe placed(hybrid) 로드가 prompt를 8행 pass로 처리해 느림
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 20/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- rust
- Domain
- backend, performance
Research direction
Read crates/gpu/src/arch/qwen3moe/body35.rs around prefill_plan and placed_open, then compare with the long-prompt GEMM batching path in body38.rs. The issue reports that placed Qwen3.8-35B-A3B prefill uses 8-row passes; done means improving that path’s prefill speed while preserving correct output. The issue says the maintainer is handling it.
Written by the indexing model from the issue text.
Description
expert 일부가 호스트에 있는 placed(hybrid) 로드에서 qwen35moe 파일의 prefill이 prompt를 8행씩 끊어 처리합니다. 12 GB RTX 3060 + 호스트 27 GB, Qwen3.8-35B-A3B Q4_K_M, 기본값으로 1464 토큰 prompt를 넣었을 때 prefill이 145 tok/s(TTFT 10.1 s)였습니다(2026-10-11, 우리 머신에서 측정).
- 원인: placed 로드에서는 GEMM 경로가 거부되고
PrefillPath::Pass로 갑니다(crates/gpu/src/arch/qwen3moe/body35.rs:1912-1918).placed_open의ubatch는MAX_PASS_ROWS = 8입니다(body35.rs:158-160,crates/gpu/src/model.rs:277). - 같은 엔진의 Qwen3.8(qwen4exp) 경로는 긴 prompt에서 GEMM ubatch를 씁니다(
body38.rs:340-341). qwen35moe의 placed 경로만 남은 것입니다. - 고칠 곳:
body35.rs의prefill_plan(1907–1921행)과placed_open.
메인테이너가 고칩니다.
- Dominant language
- Rust
- Stars
- 21
- Forks
- 4
- Avg merge
- 3h 27m
- Merged PRs (30d)
- 194
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from midagedev/bloomery
-
review-bot
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
midagedev/bloomery#318 · 1 comment ·
Maintainers usually reply within 1 day
-
review-bot
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
Maintainers usually reply within 1 day
-
review-bot
Difficulty 2/5 Under an hour Newbie friendliness 82/100
Maintainers usually reply within 1 day
-
review-bot
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
midagedev/bloomery#265 · 1 comment ·
Maintainers usually reply within 1 day
-
review-bot
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
Maintainers usually reply within 1 day
All issues in midagedev/bloomery
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
nannou-org/egui_graph#92 ·
Maintainers usually reply within 1 day
-
documentation good first issue
Difficulty 2/5 Half a day Newbie friendliness 72/100
sujithputta02/Esapay#13 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
tursodatabase/turso#10002 ·
Maintainers usually reply within 1 day
-
needs triage
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
topgrade-rs/topgrade#2419 ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 88/100
Chia-Network/clvm_rs#863 ·
Maintainers usually reply within 1 day