Repository navigation

feat: support GLM-5.3 Flash dynamic FP8 KV cache with regression cove… - #1621

Open

blueswhen wants to merge 5 commits into

mainfrom

Open

feat: support GLM-5.3 Flash dynamic FP8 KV cache with regression cove…#1621
blueswhen wants to merge 5 commits into
mainfrom
glm5.3_fp8kv

Commits on Oct 26, 2026
- feat: support GLM-5.3 Flash dynamic FP8 KV cache with regression coverage
  niushengxiao
  committed
  d9044fe
- perf: optimize FP8 KV attention and cache writes
  niushengxiao
  committed
  6e84a4a
- feat: move kernel to triton_kernel folder
  niushengxiao
  committed
  8453aea
- refactor: reuse backend FP8 KV workspaces and derive query padding dimensions
  niushengxiao
  committed
  5b8cd8b
Commits on Oct 27, 2026
- fix: avoid redundant JIT specialization in GLM FP8 KV and TP MoE kernels
  niushengxiao
  committed
  66d4988