feat(dpa4): multiple updates for dpa4 - #5734

Merged

OutisLi merged 15 commits into

deepmodeling:masterfrom

OutisLi:pr/dpa4update

Jul 7, 2026

Commits on Jul 5, 2026

perf(dpa4): add gpu native compute_edges_gpu inference
OutisLi
committed
4ca27ce
feat(pt/dpa4): allow n_blocks to be 0
OutisLi
committed
527d638
feat(pt/dpa4): add native DPA4-Spin
OutisLi
committed
50c4c17
perf(pt/dpa4): add flash attention
OutisLi
committed
65f125f
perf(pt/dpa4): layout opt
OutisLi
committed
7e6499d
perf(pt/dpa4): triton gemm
OutisLi
committed
9612262
perf(pt/dpa4): add toy CuTe realization
OutisLi
committed
a3077c2
perf(pt/dpa4): triton fusion kernel
OutisLi
committed
e1feff1
refactor(dpmodel/dpa4): align with pt backend
OutisLi
committed
2962acb
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
0d54235
fix(dpa4): pr bugs
OutisLi
committed
2dac033
perf(dpa4): c++ infer path
OutisLi
committed
b4b22df
fix
OutisLi
committed
70f4a1e

Commits on Jul 6, 2026

fix
OutisLi
committed
2ba9c0f
fix
OutisLi
committed
0ec6c5f