feat: use FIA for qwen3.5 prefill attention on npu.#1448

Open

Sinle4Cat wants to merge 10 commits into

jd-opensource:preview/qwen3.5-qwen3.6from

Sinle4Cat:x_flash

Commits on May 11, 2026

feat(npu): use FIA for qwen3.5 prefill attention
default
committed

Commits on May 13, 2026

Support x flash attention prefill on NPU
default
committed

Commits on May 14, 2026

bugfix: x flash attention long kv x mte excepiton
sinle4cat
committed
Merge preview/qwen3.5-qwen3.6 into x_flash
sinle4cat
committed

Commits on May 15, 2026

bugfix: route chunked prefill to x flash attention
Sinle4Cat
committed

Commits on May 16, 2026

bugfix: improve x flash attention long kv stability
Sinle4Cat
committed

Commits on May 18, 2026

Merge remote-tracking branch 'origin/preview/qwen3.5-qwen3.6' into pr1448-check
Sinle4Cat
committed

Commits on May 19, 2026

Merge remote-tracking branch 'origin/preview/qwen3.5-qwen3.6' into pr1448-check
Sinle4Cat
committed

Commits on May 20, 2026

bugfix: fix long kv xflash prefill grouping
Sinle4Cat
committed

Commits on May 21, 2026

chore: remove qwen35 gdn debug logging
Sinle4Cat
committed