- 27 Nov, 2025 15 commits
-
-
zhuwenwen authored
fix: moe quant bug and attention bug See merge request dcutoolkit/deeplearing/vllm!275
-
wujl5 authored
-
-
zhuwenwen authored
-
wujl5 authored
-
zhuwenwen authored
去掉宽松mtp中的隐式同步 See merge request dcutoolkit/deeplearing/vllm!274
-
王敏 authored
-
zhuwenwen authored
-
zhuwenwen authored
-
laibao authored
-
zhuwenwen authored
[fix]解决宽松mtp1报错 See merge request dcutoolkit/deeplearing/vllm!272
-
王敏 authored
-
zhuwenwen authored
-
zhuwenwen authored
-
laibao authored
-
- 26 Nov, 2025 9 commits
-
-
zhuwenwen authored
Fix blaslt miss bias. See merge request dcutoolkit/deeplearing/vllm!270
-
wanglong3 authored
-
zhuwenwen authored
[feat]支持宽松mtp See merge request dcutoolkit/deeplearing/vllm!269
-
王敏 authored
-
王敏 authored
-
zhuwenwen authored
[pref]: DS_v2_w8a8模型融掉moe.quant See merge request dcutoolkit/deeplearing/vllm!268
-
wujl5 authored
-
zhuwenwen authored
-
zhuwenwen authored
-
- 23 Nov, 2025 2 commits
- 21 Nov, 2025 9 commits
-
-
zhuwenwen authored
deepseek_v2_w8a8 增加 silu_mul_quant融合 See merge request dcutoolkit/deeplearing/vllm!265
-
wujl5 authored
-
zhuwenwen authored
set VLLM_USE_LIGHTOP_FILL_MOE_ALIGN=1, VLLM_USE_OPT_ZEROS=1 and VLLM_USE_PP_SYNC=1
-
zhuwenwen authored
feat: pp mtp加入零消耗调度,加入环境变量VLLM_USE_ZERO_MTP,默认打开 See merge request dcutoolkit/deeplearing/vllm!264
-
jujl1 authored
-
zhuwenwen authored
-
zhuwenwen authored
-
-
zhuwenwen authored
-
- 20 Nov, 2025 5 commits