[Bugfix][Quark] Fix W8A8 INT8 garbage outputs on Step-3.5-Flash (and other 3-key fused-MoE Quark exports) (#41892)

Signed-off-by: kangletian <kangletian@hotmail.com>
This commit is contained in:
PikaPikachu
2026-05-13 11:59:49 +00:00
committed by GitHub
parent 3c413a5481
commit 3b1ef03be4
3 changed files with 46 additions and 4 deletions
@@ -47,6 +47,17 @@ class QuarkW8A8Int8(QuarkScheme):
):
layer.logical_widths = output_partition_sizes
# Quark stores per-channel weight_scale as 1D [N]; reshape to [N, 1].
def _scale_weight_loader(
param: torch.nn.Parameter,
loaded_weight: torch.Tensor,
*args,
**kwargs,
):
if loaded_weight.dim() == 1:
loaded_weight = loaded_weight.unsqueeze(-1)
return weight_loader(param, loaded_weight, *args, **kwargs)
self.kernel = init_int8_linear_kernel(
is_channelwise=(self.qscheme == "per_channel"),
is_static_input_scheme=(self.is_static_input_scheme is True),
@@ -69,15 +80,15 @@ class QuarkW8A8Int8(QuarkScheme):
# WEIGHT SCALE
if self.qscheme == "per_channel":
weight_scale = ChannelQuantScaleParameter(
data=torch.empty((sum(output_partition_sizes)), dtype=torch.float32),
data=torch.empty((sum(output_partition_sizes), 1), dtype=torch.float32),
output_dim=0,
weight_loader=weight_loader,
weight_loader=_scale_weight_loader,
)
ChannelQuantZPParameter = ChannelQuantScaleParameter
weight_zero_point = ChannelQuantZPParameter(
data=torch.empty((sum(output_partition_sizes)), dtype=torch.int8),
data=torch.empty((sum(output_partition_sizes), 1), dtype=torch.int8),
output_dim=0,
weight_loader=weight_loader,
weight_loader=_scale_weight_loader,
)
else:
assert self.qscheme == "per_tensor"