Repository navigation

[Feature] Support layerwise KV pool for Qwen3.5 hybrid GDN+full_attention models - #12711

Merged

kunpengW-code merged 5 commits into

vllm-project:mainfrom

Sep 2, 2026

Commits on Jul 22, 2026
- feat(qwen3.5): support layerwise KV pool for hybrid GDN+full_attention models
  tyy0829
  committed
  ab384eb
Commits on Aug 21, 2026
- test(gdn): stub NPU side-effects in layerwise KV compile test
  tyy0829
  committed
  10a625b
Commits on Aug 26, 2026
- fix(kv-pool): only trim external hit for Eagle when it reaches the final block
  tyy0829
  committed
  7bc67af
Commits on Aug 28, 2026
- fix(kv-transfer): order Mamba state copy after layer load
  tyy0829
  committed
  793727f
Commits on Aug 31, 2026
- docs: explain the layerwise mamba-copy ordering in model_runner_v1
  tyy0829
  committed
  807a9cb