v2.2.5

What's Changed

• Support Single Node P-D Disaggregate function for CUDA.
• Support Qwen3 Model, Currently only for Dense, MoE Models support is WIP.
• Support EP in MOE OP, currently only support BF16 and FP16.
• CPU Support is not work in this release, Use a v2.1.x version for CPU support.

In Detail

• readme: add citation, update subproject description by @leefige in #70
• fix cugraph dnn moe kernel bug by @laiwenzh in #71
• support MOE EP by @yjc9696 in #73
• support disaggregated prefilling by @laiwenzh in #78
• prompt相关的问题修复 by @lddfym in #77
• Build: only build flash attention kernel once. by @kzjeef in #80
• support qwenV3【Dense】 by @yjc9696 in #81
• model: fix cpu compiler error by @kzjeef in #82
• Fix cpu compile by @kzjeef in #83
• Create build-check.yml by @kzjeef in #84
• ci: only trigger cuda release for current version. by @kzjeef in #85

New Contributors

• @yjc9696 made their first contribution in #73
• @lddfym made their first contribution in #77

Full Changelog: v2.1.0...v3.0.0-rc1

Existing Issue

• CPU Support is not work in this release, Use a v2.1.x version for CPU support.