Skip to content

Metal backend: fix linear_filter in test_modules #17253

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
manuelcandales merged 3 commits into from
Feb 5, 2026
Merged

Metal backend: fix linear_filter in test_modules #17253

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
401af46
Update
manuelcandales Feb 5, 2026
87f1529
Update
manuelcandales Feb 5, 2026
cf89a2b
Update
manuelcandales Feb 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 12 additions & 10 deletions backends/apple/metal/tests/test_modules.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -287,15 +287,15 @@ def forward(self, x: torch.Tensor):
class LinearInt4_QMV_IMPL_small_odd(nn.Module):
def __init__(self):
super().__init__()
self.linear = nn.Linear(8, 3, bias=True)
self.linear = nn.Linear(32, 3, bias=True)

def forward(self, x: torch.Tensor):
return self.linear(x)


MODULE_REGISTRY["linear_int4_qmv_impl_small_odd"] = {
"model_class": LinearInt4_QMV_IMPL_small_odd,
"input_shapes": [(1, 8)],
"input_shapes": [(1, 32)],
"description": "Linear int4 quantization dispatching to qmv_impl",
"qlinear": "fpa4w",
"qlinear_group_size": 32,
Expand All @@ -312,15 +312,15 @@ def forward(self, x: torch.Tensor):
class LinearInt4_QMV_IMPL_small_even(nn.Module):
def __init__(self):
super().__init__()
self.linear = nn.Linear(8, 10, bias=True)
self.linear = nn.Linear(32, 10, bias=True)

def forward(self, x: torch.Tensor):
return self.linear(x)


MODULE_REGISTRY["linear_int4_qmv_impl_small_even"] = {
"model_class": LinearInt4_QMV_IMPL_small_even,
"input_shapes": [(1, 8)],
"input_shapes": [(1, 32)],
"description": "Linear int4 quantization dispatching to qmv_impl",
"qlinear": "fpa4w",
"qlinear_group_size": 32,
Expand Down Expand Up @@ -694,12 +694,14 @@ def quantize_model(model: nn.Module, qlinear: str, qlinear_group_size: int = 32)
else:
raise ValueError(f"Unsupported linear quantization config '{qlinear}'.")

def linear_filter(module, fqn):
if isinstance(module, torch.nn.Linear):
# Check if hidden dimension is divisible by group size
return qlinear_group_size == 0 or (
module.weight.shape[1] % qlinear_group_size == 0
)
def linear_filter(m, fqn):
if isinstance(m, torch.nn.Linear):
if m.weight.shape[1] % qlinear_group_size != 0:
raise ValueError(
f"Metal int4 quantization requires weight dimension (K) to be multiple of group_size. "
f"Layer {fqn} has weight shape {m.weight.shape} (K={m.weight.shape[1]}, group_size={qlinear_group_size})" # noqa: E501
)
return True
return False

quantize_(model, linear_config, filter_fn=linear_filter)
Expand Down
Loading