[ROCm][Quantization][5/N] Refactor quark_moe w8a8-int8 w/ oracle (#46765)

Signed-off-by: amd-sourjya <amd-sourjya@users.noreply.github.com>
Co-authored-by: amd-sourjya <amd-sourjya@users.noreply.github.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
amd-sourjya
2026-07-27 16:01:34 -05:00
committed by GitHub
co-authored by amd-sourjya Cursor Andreas Karatzas
parent b5bcb3ce88
commit 1053e248f0
8 changed files with 234 additions and 17 deletions
@@ -190,6 +190,8 @@ kInt4Static32Asym = QuantKey(
kInt8StaticChannelSym = QuantKey(torch.int8, kStaticChannelScale, symmetric=True)
kInt8DynamicTokenSym = QuantKey(torch.int8, kDynamicTokenScale, symmetric=True)
kInt8StaticTensorSym = QuantKey(torch.int8, kStaticTensorScale, symmetric=True)
kInt8DynamicTensorSym = QuantKey(torch.int8, kDynamicTensorScale, symmetric=True)
# INT4 W4A8 quantization keys