fix: GLM-5.2 w/ DCP+DSpark for SM120 - #1765

Merged

AlpinDale merged 11 commits into

mainfrom

fix/sm120-glm-dspark-dcp

Aug 13, 2026

Commits on Aug 12, 2026

fix(spec-decode): isolate packed MLA draft cache
AlpinDale
committed
af0bc7c
feat(cuda): add native SM120 DSA indexer
AlpinDale
committed
b4114b5
fix(runtime): isolate Triton caches per worker
AlpinDale
committed
4b1c1da
fix(cuda): restore SM120 sparse MLA prefill
AlpinDale
committed
2802232
fix(attention): harden sparse MLA DCP decode
AlpinDale
committed
5911698
fix(attention): align sparse indexer block tables
AlpinDale
committed
ba3741e

Commits on Aug 13, 2026

fix(moe): make shared expert overlap graph safe
AlpinDale
committed
542ec47
fix(spec-decode): support DSpark under DCP
AlpinDale
committed
7c9c459
fix(cuda): read SM120 indexer cache layout correctly
AlpinDale
committed
7bc8c35
fix(attention): handle fused GLM decode under DCP
AlpinDale
committed
fcfa04b
fix(spec-decode): type draft attention metadata
AlpinDale
committed
36bd977