auto-round
211e95f8 - perf(ark): decode int4-sym nibbles through the asym unsigned path (^0x88 sign-flip)

Commit

29 days ago

perf(ark): decode int4-sym nibbles through the asym unsigned path (^0x88 sign-flip) Co-authored-by: a32543254 <53296245+a32543254@users.noreply.github.com>

References

#2111 - Optimize INT4 MoE prefill/decode (sym) XPU kernels with dedicated w4a16 tiling

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

Author

Copilot

Parents

1f7b6313

auto-round 211e95f8 - perf(ark): decode int4-sym nibbles through the asym unsigned path (^0x88 sign-flip)

auto-round
211e95f8 - perf(ark): decode int4-sym nibbles through the asym unsigned path (^0x88 sign-flip)