Add CUDA graph capture/replay for qwen 3.5 moe decode method#18809

Open

Gasoonjia wants to merge 23 commits intomainfrom

Commits on Apr 2, 2026

Replace chunked FLA with recurrent gated delta rule for T=1 decode
Gasoonjia
committed

Commits on Apr 5, 2026

Dual-method PTE with GPU-resident state for Qwen3.5 MoE
gasoonjia
committed

Commits on Apr 6, 2026

Use share_mutable_buffers to eliminate select_scatter overhead
gasoonjia
committed
Merge branch 'main' into recurrent-fla

Gasoonjia
authored and
gasoonjia
committed
lint
Gasoonjia
committed
remove reduntdant updates
Gasoonjia
committed

Commits on Apr 7, 2026

Commits on Apr 9, 2026

Improve CUDA backend error handling and add dual-method runner fallback
gasoonjia
committed

Commits on Apr 10, 2026

Commits on Apr 13, 2026