auto-round
35e89aee - fix(ark): carry the 32-byte W4A8 decode chunk as 16-byte sycl::vec sub-vectors

Commit

8 days ago

fix(ark): carry the 32-byte W4A8 decode chunk as 16-byte sycl::vec sub-vectors Co-authored-by: a32543254 <53296245+a32543254@users.noreply.github.com>

References

#2143 - feat: W4A8 ARK XPU MoE kernel (int4 weight / int8 compute) with prefill + decode

Author

Copilot

Parents

725b2c82

auto-round 35e89aee - fix(ark): carry the 32-byte W4A8 decode chunk as 16-byte sycl::vec sub-vectors

auto-round
35e89aee - fix(ark): carry the 32-byte W4A8 decode chunk as 16-byte sycl::vec sub-vectors