WINT4/WINT8 dense gemm default use Machete (#4451)

This commit is contained in:
Sunny-bot1
2025-10-23 17:57:59 +08:00
committed by GitHub
parent a240425db9
commit 4ffe41a747
12 changed files with 310 additions and 15 deletions
@@ -167,7 +167,7 @@ def machete_quantize_and_pack(
atype,
quant_type,
scale_type,
)[0]
)
return w_q_prepack, w_s
@@ -194,5 +194,5 @@ def machete_wint_mm(
out_dtype, # out_dtype
group_size, # group_size
scheduler, # scheduler
)[0]
)
return out