- 14 Apr, 2026 3 commits
- 10 Apr, 2026 3 commits
- 08 Apr, 2026 2 commits
- 01 Apr, 2026 1 commit
-
-
laibao authored
-
- 27 Mar, 2026 2 commits
-
-
flyingdown authored
-
flyingdown authored
-
- 26 Mar, 2026 2 commits
- 24 Mar, 2026 4 commits
- 21 Mar, 2026 4 commits
- 20 Mar, 2026 1 commit
-
-
laibao authored
-
- 19 Mar, 2026 1 commit
-
-
laibao authored
移除 forward 中对 experts.use_overlapped/_shared_experts 的状态改写,避免 torch.compile 启动期 shared/non-shared 路径不一致 FusedMoE.forward_impl 仅在 shared_output 为空时计算 shared experts,防止透传值被本地重算覆盖
-
- 18 Mar, 2026 3 commits
-
-
laibao authored
新增环境变量 VLLM_USE_LIGHTOP_MOE_SUM_MUL_ADD 用于控制 fused sum+mul+add 开关。 在 DeepseekV2MoE 中增加 fused 路径,预计算 shared_output,并下传 iqis 与 routed_scaling_factor。 扩展 FusedMoE/SharedFusedMoE 及相关 custom op 接口,统一透传 i_q/i_s/shared_output/routed_scaling_factor。 同步适配 Triton、Marlin W16A16、SlimQuant W4A8、CompressedTensors W8A8 等实现,支持在内核侧完成 sum+mul+add。
-
lixh6 authored
-
yangql authored
-
- 16 Mar, 2026 2 commits
- 13 Mar, 2026 3 commits
- 12 Mar, 2026 8 commits
- 11 Mar, 2026 1 commit
-
-
liuchy5 authored
-