[WIP] feat(lora): LoRA adapter serving#83

Draft

qywu wants to merge 46 commits into

lightseekorg:mainfrom

qywu:feat/lora-adapter-serving

Commits on May 7, 2026

Commits on May 8, 2026

Commits on May 11, 2026

Merge remote-tracking branch 'upstream/main' into feat/lora-adapter-serving
qywu
committed

Commits on May 13, 2026

Commits on May 14, 2026

Commits on May 18, 2026

Commits on May 19, 2026

refactor(lora): rename chunked_sgmv_expand to lora_expand_prefill
qywu
committed
perf(lora): add lora_shrink_prefill and dispatch shrink on max_len
qywu
committed
build(lora): add comprehensive autotune sweep script
qywu
committed
perf(lora): populate autotune caches for common model shapes on H100
qywu
committed
perf(lora): kernel micro-optimisations in decode shrink/expand
qywu
committed
perf(lora): grouped decode expand for tensor-core efficiency
qywu
committed
perf(lora): refresh autotune caches after decode kernel micro-opts
qywu
committed
perf(lora): expand configs BLOCK_N=128 + BLOCK_K=64 from profiling
qywu
committed
perf(lora): eliminate k-mask via tl.multiple_of across all decode kernels
qywu
committed
perf(lora): vLLM-style adapter-grouped expand without gather/scatter
qywu
committed
fix(lora): restore k-mask in grpv2 to prevent BLOCK_K > rank silent miscompute
qywu
committed
Merge remote-tracking branch 'upstream/main' into pr-83-resolve
qywu
committed
perf(lora): adapter-grouped expand + correctness fix + benchmarks
qywu
committed
perf(lora): unify seg+grpv2 via max_group_size grid — no dispatch threshold needed
qywu
committed

Commits on May 20, 2026