Repository navigation

fix(gemma4): drive the TP ranks concurrently so long prompts do not stall - #1

Closed

sysuTobo wants to merge 7 commits into

feat/gemma4-tp2from

feat/gemma4-tp2-concurrent

Closed

fix(gemma4): drive the TP ranks concurrently so long prompts do not stall#1
sysuTobo wants to merge 7 commits into
feat/gemma4-tp2from
feat/gemma4-tp2-concurrent

Commits on Oct 3, 2026
Commits on Oct 4, 2026
- refactor(kernels): map CUDA and cuBLAS errors through one shared table (#1126)
  FeathBow
  authored
  25672c1
Commits on Oct 6, 2026
- feat(kernels): make the gated RMSNorm activation selectable (#1129)
  sysuTobo
  authored
  c8a09c9
- feat(gemma4): serve Gemma 4 31B bf16 at tensor parallel 2 (#1115)
  sysuTobo
  authored
  ee4672c
Commits on Oct 7, 2026
- fix(gemma4): drive the TP ranks concurrently so long prompts do not stall
  sysuTobo
  committed
  9d000cb