[FEAT] 4 unimplemented optimizations from PR391 plus 2 new ones - six serving optimizations for Qwen3.8 Flash-Next on MTPLX 2.11.2 (four PR-391 remainder optimizations + two exact decode optimizations) - #475

Open

davidtai wants to merge 11 commits into

youssofal:mainfrom

davidtai:perf/qwen38-aux-lanes

Commits on Sep 7, 2026

perf(qwen4): three PR 391 remainder lanes for Qwen3.8 Flash-Next onto 2.11.2
davidtai
committed
3201b21
perf(qwen4): QSA split-K sparse-GQA decode lane for Qwen3.8 Flash-Next
davidtai
committed
2549733
fix(qwen4): arm MTPLX_QSA_SPARSE_DECODE at install time, not import
davidtai
committed
64e8164
fix(qwen4): resolve every remainder-lane flag at use, not at import
davidtai
committed
16997fd
fix(qwen4): resolve the four frozen-at-import verify-lane flags at use
davidtai
committed
5dd05b7
feat(qwen4): surface the three no-observable verify lanes in /health
davidtai
committed
3729c09
perf(qwen4): two exact decode lanes for Qwen3.8 Flash-Next, on the PR 391 remainder port
davidtai
committed
27d5ff6
docs(perf): Qwen3.8 Flash-Next battery for the six serving optimizations
davidtai
committed
038b2b0
docs(perf): regenerate the battery charts without superseded windows; add the 16K interleave chart
davidtai
committed
aab2aca

Commits on Sep 9, 2026