- 15 Jan, 2026 1 commit
-
-
wanglong3 authored
-
- 12 Jan, 2026 2 commits
- 09 Jan, 2026 2 commits
- 08 Jan, 2026 2 commits
- 07 Jan, 2026 1 commit
-
-
wujl5 authored
-
- 06 Jan, 2026 1 commit
-
-
wanglong3 authored
-
- 05 Jan, 2026 3 commits
- 04 Jan, 2026 1 commit
-
-
wujl5 authored
-
- 25 Dec, 2025 1 commit
-
-
gaoqiong authored
-
- 23 Dec, 2025 5 commits
- 22 Dec, 2025 5 commits
- 21 Dec, 2025 1 commit
-
-
yangql authored
-
- 20 Dec, 2025 1 commit
-
-
zhuwenwen authored
-
- 18 Dec, 2025 4 commits
- 17 Dec, 2025 3 commits
- 16 Dec, 2025 4 commits
-
-
zhuwenwen authored
-
yangql authored
-
zhuwenwen authored
set VLLM_USE_LIGHTOP_RMS_ROPE_CONCAT=1
-
laibao authored
移除 fuse_moe_w16a16_marlin.py 中未被调用的量化准备逻辑与相关导入 精简 fused_experts_impl_w16a16_marlin 的未用量化参数,并同步更新 fused_moe.py 调用点 清理 marlin_quant.py 中未被 vLLM 主路径使用的函数与 __main__ 调试入口,仅保留 W16A16 权重打包所需实现 添加MOE_SUM_MUL_ADD、MOE_SUM优化
-
- 15 Dec, 2025 3 commits