[CUDA] QMoE GEMV fast path for batch-1 decode #29038
QMoE GEMV kernel
a4474774
address feedbacks
a37ac011
gemv for group wise quantization
c51e7b4a
support bf16
9f701737
support int8 per-column
b9c35807
tianleiwu
changed the title Add INT4 per-channel MoE GEMV fast path for batch-1 decode QMoE GEMV fast path for batch-1 decode 38 days ago
tianleiwu
changed the title QMoE GEMV fast path for batch-1 decode [CUDA] QMoE GEMV fast path for batch-1 decode 38 days ago
support block size 32
a8088b9d
Fix buffer aliasing and Use fp32 accumulation
f7b900e8
lintrunner
79cb66ad
tianleiwu
merged
56f6fee6
into main 37 days ago
tianleiwu
deleted the tlwu/20260613/qmoe_gemv branch 37 days ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub