- 11 Feb, 2026 1 commit
-
-
zhuwenwen authored
fuse_moe_fp8接入marlin算子 fix(v1):修复抢占恢复时 BlockTable 溢出 feat(moe):新增 VLLM_USE_MOE_W16A16_TRTION 强制 Triton MoE fix: 解决原版0消耗chunk-prefill崩溃问题 fp8增加fused_moe_gate参数
-
- 10 Feb, 2026 3 commits
-
-
zhuwenwen authored
[opt] 优化epsp代码, 零消耗添加epsp update VLLM_USE_FUSED_RMS_ROPE=0 (default). for qwen3, VLLM_USE_FUSED_RMS_ROPE=1 (default) feat(moe/marlin): Marlin W16A16 MoE 自动探测并预打包(去掉手动开关) perf(qwen3): 融合 q/k RMSNorm + RoPE fused_moe_fp8接入lmslim
-
zhuwenwen authored
[qwen3-480b] MoE(TN) configs for nmz TP=8 [opt] 优化deepep相关代码 [fix] 修复deepseek moe模型的awq量化推理bug和精度问题, 修复awq模型的VLLM_USE_LIGHTOP_MOE_SUM_MUL_ADD设置位置, update_state,优化性能,去除冗余操作 pcie 解决custom cudagraph模式需要拷贝的问题,需要配合dtk进行使用 [feat] Switch default w8a8 gemm impl to blaslt. Support w8a8-fp8 GEMM backend.MoE 路由抓取:新增 router_capture 工具链与 envs 统一配置 [envs] set VLLM_CUSTOM_CACHE=1、VLLM_USE_FUSED_RMS_ROPE=1、VLLM_USE_FUSED_FILL_RMS_CAT=1、VLLM_USE_FLASH_ATTN_FP8=1、VLLM_USE_FLASH_MLA_FP8=1、update VLLM_USE_TOPK_RENORM
-
zhuwenwen authored
[pd] 修改非堆成切分的判断, pd分离使用默认调度,set VLLM_USE_PD_SPLIT=1 [perf](fused-moe): 预打包 Marlin W16A16 MoE 权重,降低 warmup 显存峰值 [fix]修复丢弃MTP代码报错,pp+chunkprefill多并发input ids更新bug, 修复不开启融合图的断言错误, PP 场景 decode 阶段 token 被误丢弃导致卡住
-
- 09 Feb, 2026 3 commits
- 06 Jan, 2026 1 commit
-
-
jujl1 authored
-
- 04 Jan, 2026 2 commits
- 31 Dec, 2025 3 commits
- 29 Dec, 2025 1 commit
-
-
yangql authored
-
- 26 Dec, 2025 1 commit
-
-
zhuwenwen authored
-
- 25 Dec, 2025 2 commits
- 24 Dec, 2025 3 commits
- 23 Dec, 2025 2 commits
- 22 Dec, 2025 2 commits
- 21 Dec, 2025 1 commit
-
-
yangql authored
-
- 19 Dec, 2025 2 commits
- 18 Dec, 2025 4 commits
- 17 Dec, 2025 4 commits
- 16 Dec, 2025 2 commits
- 09 Dec, 2025 1 commit
-
-
王敏 authored
-
- 08 Dec, 2025 2 commits