Skip to content

[xpu][test] Port more cases in test_nf4.py and test_quant_api.py UT files to intel XPU #3475

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

Open
zxd1997066 wants to merge 3 commits into
base: main
Choose a base branch
from
Open

[xpu][test] Port more cases in test_nf4.py and test_quant_api.py UT files to intel XPU #3475

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
7ee6a90
port more case in test_nf4.py to xpu
zxd1997066 Dec 10, 2025
3b1ecf9
port more case in test_quant_api.py to xpu
zxd1997066 Dec 10, 2025
d01032d
skip nf4 fsdp2
zxd1997066 Dec 12, 2025
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
2 changes: 1 addition & 1 deletion test/dtypes/test_nf4.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -262,7 +262,7 @@ def test_smoketest_linear(self, dtype: torch.dtype):
_ = torch.nn.functional.linear(inp, a)
_ = torch.nn.functional.linear(inp, a_nf4)

@unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")
@unittest.skipIf(not torch.accelerator.is_available(), "Need GPU available")
@parametrize("dtype", [torch.bfloat16, torch.float16, torch.float32])
def test_smoketest_linear_compile(self, dtype: torch.dtype):
if (
Expand Down
22 changes: 11 additions & 11 deletions test/quantization/test_quant_api.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -643,20 +643,20 @@ def test_module_fqn_to_config_module_name(self):
assert isinstance(model.linear2.weight, AffineQuantizedTensor)
assert isinstance(model.linear2.weight._layout, PlainLayout)

@unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")
@unittest.skipIf(not torch.accelerator.is_available(), "Need GPU available")
def test_module_fqn_to_config_regex_basic(self):
config1 = Int4WeightOnlyConfig(
group_size=32, int4_packing_format="tile_packed_to_4d"
)
config = ModuleFqnToConfig({"re:linear.": config1})
model = ToyLinearModel().cuda().to(dtype=torch.bfloat16)
example_inputs = model.example_inputs(device="cuda", dtype=torch.bfloat16)
model = ToyLinearModel().to(_DEVICE).to(dtype=torch.bfloat16)
example_inputs = model.example_inputs(device=_DEVICE, dtype=torch.bfloat16)
quantize_(model, config, filter_fn=None)
model(*example_inputs)
assert isinstance(model.linear1.weight, Int4TilePackedTo4dTensor)
assert isinstance(model.linear2.weight, Int4TilePackedTo4dTensor)

@unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")
@unittest.skipIf(not torch.accelerator.is_available(), "Need GPU available")
def test_module_fqn_to_config_regex_precedence(self):
"""Testing that full path config takes precedence over
regex config in ModuleFqnToConfig
Expand All @@ -666,14 +666,14 @@ def test_module_fqn_to_config_regex_precedence(self):
)
config2 = IntxWeightOnlyConfig()
config = ModuleFqnToConfig({"linear1": config1, "re:linear.": config2})
model = ToyLinearModel().cuda().to(dtype=torch.bfloat16)
example_inputs = model.example_inputs(device="cuda", dtype=torch.bfloat16)
model = ToyLinearModel().to(_DEVICE).to(dtype=torch.bfloat16)
example_inputs = model.example_inputs(device=_DEVICE, dtype=torch.bfloat16)
quantize_(model, config, filter_fn=None)
model(*example_inputs)
assert isinstance(model.linear1.weight, Int4TilePackedTo4dTensor)
assert isinstance(model.linear2.weight, IntxUnpackedToInt8Tensor)

@unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")
@unittest.skipIf(not torch.accelerator.is_available(), "Need GPU available")
def test_module_fqn_to_config_regex_precedence2(self):
"""Testing that full path config takes precedence over
regex config in ModuleFqnToConfig, swapping
Expand All @@ -685,14 +685,14 @@ def test_module_fqn_to_config_regex_precedence2(self):
)
config2 = IntxWeightOnlyConfig()
config = ModuleFqnToConfig({"re:linear.": config2, "linear1": config1})
model = ToyLinearModel().cuda().to(dtype=torch.bfloat16)
example_inputs = model.example_inputs(device="cuda", dtype=torch.bfloat16)
model = ToyLinearModel().to(_DEVICE).to(dtype=torch.bfloat16)
example_inputs = model.example_inputs(device=_DEVICE, dtype=torch.bfloat16)
quantize_(model, config, filter_fn=None)
model(*example_inputs)
assert isinstance(model.linear1.weight, Int4TilePackedTo4dTensor)
assert isinstance(model.linear2.weight, IntxUnpackedToInt8Tensor)

@unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")
@unittest.skipIf(not torch.accelerator.is_available(), "Need GPU available")
def test_module_fqn_to_config_regex_fullmatch(self):
"""Testing that we will only match the fqns that fully
matches the regex
Expand Down Expand Up @@ -731,7 +731,7 @@ def example_inputs(self):
"linear3_full_match.bias": None,
}
)
model = M(dtype=torch.bfloat16, device="cuda")
model = M(dtype=torch.bfloat16, device=_DEVICE)
example_inputs = model.example_inputs()
quantize_(model, config, filter_fn=None)
model(*example_inputs)
Expand Down
2 changes: 1 addition & 1 deletion torchao/quantization/quantize_/workflows/int4/int4_tile_packed_to_4d_tensor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,7 +120,7 @@ def from_hp(
# Validate kernel requirements
orig_out_features, orig_in_features = hp_tensor.shape[-2:]
# TODO: relax checks to enable quantizing in other platoforms and run in A100
if not torch.cuda.get_device_capability()[0] >= 8:
if torch.cuda.is_available() and not torch.cuda.get_device_capability()[0] >= 8:
raise ValueError(
f"Cannot use tinygemm int4 kernel with a device of compute capability {torch.cuda.get_device_capability()}, the minimum compute capability is 8.0 for tensor core kernels."
)
Expand Down