Skip to content

fix: use Int4WeightOnlyConfig.group_size in fake quant configs #4518

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
vkuzo merged 1 commit into from
Jul 7, 2026
Merged

fix: use Int4WeightOnlyConfig.group_size in fake quant configs #4518

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
7 changes: 4 additions & 3 deletions test/quantization/test_qat.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1951,19 +1951,20 @@ def test_infer_fp8_int4_config(self):
self.assertEqual(weight_config.group_size, 128)
self.assertEqual(weight_config.activation_dtype, e4m3_dtype)

def test_infer_int4_weight_only_config(self):
@parametrize("group_size", [256, 128, 64, 32])
def test_infer_int4_weight_only_config(self, group_size: int):
"""
Test that fake quantize configs are correctly inferred from `Int4WeightOnlyConfig`.
"""
from torchao.quantization.qat.fake_quantize_config import (
_infer_fake_quantize_configs,
)

base_config = Int4WeightOnlyConfig(version=2)
base_config = Int4WeightOnlyConfig(version=2, group_size=group_size)
(act_config, weight_config) = _infer_fake_quantize_configs(base_config)
self.assertIsNone(act_config)
self.assertIsInstance(weight_config, Int4WeightFakeQuantizeConfig)
self.assertEqual(weight_config.group_size, 128)
self.assertEqual(weight_config.group_size, group_size)
self.assertEqual(weight_config.activation_dtype, torch.bfloat16)

@unittest.skipIf(not is_sm_at_least_89(), "Need sm89+")
Expand Down
2 changes: 1 addition & 1 deletion torchao/quantization/qat/fake_quantize_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -382,7 +382,7 @@ def _infer_fake_quantize_configs(
f"Packing format must be one of {supported_packing_formats}"
)
weight_config = Int4WeightFakeQuantizeConfig(
group_size=128,
group_size=base_config.group_size,
activation_dtype=torch.bfloat16,
)
else:
Expand Down
Loading