v0.9.0.2rc2

Full Changelog: v0.9.2rc1...v0.9.2rc2

v0.9.2rc1

Full Changelog: v0.8.6rc1...v0.9.2rc1

v0.8.6rc1

Full Changelog: v0.8.4rc4...v0.8.6rc1

v0.8.5rc2

What's Changed

• [Misc][VIP] ignore marlin_moe_wna16 local gen codes by @DefTruth in #27

Full Changelog: v0.8.4rc4...v0.8.5rc2

v0.8.5rc1

What's Changed

• [Bugfix][Kernel][VIP] fix potential cuda graph broken for merge_attn_states kernel by @DefTruth in #26

Full Changelog: v0.8.4rc4...v0.8.5rc1

v0.8.4rc4

Full Changelog: v0.8.4rc3...v0.8.4rc4

v0.8.4rc3

Rebase to vllm:main for a more clean codebase
Full Changelog: https://github.com/vipshop/vllm/commits/v0.8.4rc3

v0.8.4rc2

What's Changed

• [Kernel] optimize merge_attn_states CUDA kernel dispatch by @DefTruth in #22
• [Update][VIP] Update from vllm:main and fix conflicts by @DefTruth in #23
• [Kernel][VIP] opt cuda merge_attn_states kernel thread block dispatch by @DefTruth in #24

Full Changelog: v0.8.4rc1...v0.8.4rc2

v0.8.4rc1

What's Changed

• [Kernel][VIP] support cuda merge_attn_states kernel, max ~3x improved by @DefTruth in #18
• [Kernel][VIP] support cuda merge_attn_states kernel by @DefTruth in #19
• [Kernel][VIP] dispatch merge_attn_states cuda kernel, half&bf16 by @DefTruth in #20
• [Misc][VIP] Revert to original Tritron merge_attn_states kernel by @DefTruth in #21

Full Changelog: v0.8.4rc0...v0.8.4rc1

v0.8.4rc0

Full Changelog: v0.8.3rc1...v0.8.4rc0