[LoRA] MoE-LoRA CUDA-graph fixes under DP attention + RL adapter-reload fixes - #31525

Open

yushengsu-thu wants to merge 2 commits into

mainfrom

yusheng/lora-moe-cg-dp-attention

Commits on Jul 17, 2026

[LoRA] DP-attention support for LoRA: idle-forward guards, attn-TP-local slicing, per-rank tensor serialization

nanjiangwill
authored and
yushengsu-thu
committed
d81fc6e
[LoRA] MoE-LoRA cuda-graph fixes under DP-attention + RL adapter-reload fixes
yushengsu-thu
committed
150464d