Skip to content
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

Making sure int4 weight only supports cpu as well #1203

Merged
merged 1 commit into from
Oct 30, 2024
Merged

Making sure int4 weight only supports cpu as well #1203

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
23 changes: 23 additions & 0 deletions test/dtypes/test_affine_quantized.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,7 +133,30 @@ def test_print_quantized_module(self, apply_quant):
assert "AffineQuantizedTensor" in str(ql)


class TestAffineQuantizedBasic(TestCase):
COMMON_DEVICES = ["cpu"] + (["cuda"] if torch.cuda.is_available() else [])
COMMON_DTYPES = [torch.bfloat16]

@common_utils.parametrize("apply_quant", get_quantization_functions(False, True))
@common_utils.parametrize("device", COMMON_DEVICES)
@common_utils.parametrize("dtype", COMMON_DTYPES)
def test_flatten_unflatten(self, apply_quant, device, dtype):
l = torch.nn.Linear(128, 256, dtype=dtype, device=device)
ql = apply_quant(l)
lp_tensor = ql.weight
tensor_data_name_dict, tensor_attributes = lp_tensor.__tensor_flatten__()
tensor_data_dict = {name: getattr(lp_tensor, name) for name in tensor_data_name_dict}
outer_size = lp_tensor.size()
outer_stride = lp_tensor.stride()
reconstructed = type(lp_tensor).__tensor_unflatten__(tensor_data_dict, tensor_attributes, outer_size, outer_stride)
example_inputs = (torch.randn(32, 128, dtype=dtype, device=device),)
ref = ql(*example_inputs)
ql.weight = torch.nn.Parameter(reconstructed, requires_grad=False)
reconstruct_res = ql(*example_inputs)
self.assertEqual(reconstruct_res, ref)

common_utils.instantiate_parametrized_tests(TestAffineQuantized)
common_utils.instantiate_parametrized_tests(TestAffineQuantizedBasic)


if __name__ == "__main__":
Expand Down
7 changes: 5 additions & 2 deletions torchao/dtypes/affine_quantized_tensor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1268,8 +1268,11 @@ def from_plain(
def to(self, *args, **kwargs):
kwargs = self._get_to_kwargs(*args, **kwargs)
device = kwargs["device"]
if not is_device("cuda", device):
raise ValueError(f"TensorCoreTiledAQTTensorImpl is only available for cuda device, can't convert to {device}")
# tensor core tiled layout supports both cpu and cuda but does not support the conversion
# between these two devices, in the future we should not use the same layout for
# cpu and cuda device: https://github.com/pytorch/ao/issues/1117
if not is_device(torch.device(self.device).type, device):
raise ValueError(f"TensorCoreTiledAQTTensorImpl does not support conversion from {self.device} to {device}")
return self.__class__(
self.packed_weight.to(device),
self.scale_and_zero.to(device),
Expand Down
Loading