Repository navigation

Fuse shared-length query scaling during inference - #1168

Open

autumnust wants to merge 6 commits into

codex/autoresearch-cache-pinningfrom

codex/autoresearch-query-scaling-preparation

Open

Fuse shared-length query scaling during inference#1168
autumnust wants to merge 6 commits into
codex/autoresearch-cache-pinningfrom
codex/autoresearch-query-scaling-preparation

Commits on Oct 16, 2026
- Fuse inference query scaling while preserving eager rounding
  Lei Sun
  committed
  48423cb
- Apply style checks to fused query scaling
  Lei Sun
  committed
  d4b98b7
- Cover query scaling dtype and optional-kernel fallbacks
  Lei Sun
  committed
  50342fb
- Prepare shared query scales inside the fused inference kernel
  Lei Sun
  committed
  deb8e2c
- Check graph capture for fused and fallback query scaling
  Lei Sun
  committed
  834cfc5
- Keep native scaling on non-CUDA backends
  Lei Sun
  committed
  ba79370