[Perf][SM70] Add opt-in native-g32 AWQ QPN M1 dispatch - #521

Merged

yangzhuxinyzx merged 9 commits into

1CatAI:mainfrom

Leonccaa:perf/sm70-awq-qpn-m1-runtime-main-20260906

Sep 6, 2026

Commits on Sep 5, 2026

[Perf][SM70] Narrow Qwen3.8 AWQ active grouped decode
Leonccaa
committed
0cd6c80
docs: record narrow AWQ grouped decode and opt-in tuning evidence
Leonccaa
committed
96b9293
test: preserve indexed-prefill admission on aligned AWQ base
Leonccaa
committed
021ea59
docs: pin shared FP16 opt-ins and PCIe collective provenance
Leonccaa
committed
b50bb1f
docs: record aligned AWQ model integration acceptance
Leonccaa
committed
bc4686e
docs: correct engine-reported AWQ KV capacity
Leonccaa
committed
fbaa195

Commits on Sep 6, 2026

[Kernel][SM70] Add native-g32 AWQ QPN M1 operator
Leonccaa
committed
9ae1a29
[Perf][SM70] Add opt-in native-g32 AWQ QPN M1 dispatch
Leonccaa
committed
4d95867
[Docs][SM70] Define controlled numerical perturbation acceptance
Leonccaa
committed
356c1ba