Skip to content

[bug fix] Torch/IPEX 2.1.0][CI] UT issue #1477

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
merged 2 commits into from
Dec 19, 2023
Merged

[bug fix] Torch/IPEX 2.1.0][CI] UT issue #1477

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7239e2a
fix model size issue when torch=2.1
xin3he Dec 19, 2023
7690229
fix bug in recover_f_json
xin3he Dec 19, 2023
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
3 changes: 3 additions & 0 deletions neural_compressor/utils/pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -507,6 +507,9 @@ def recover_model_from_json(model, json_file_path, example_inputs):
if isinstance(example_inputs, dict):
model(**example_inputs)
model(**example_inputs)
elif isinstance(example_inputs, tuple) or isinstance(example_inputs, list):
model(*example_inputs)
model(*example_inputs)
else:
model(example_inputs)
model(example_inputs)
Expand Down
28 changes: 14 additions & 14 deletions test/model/test_model_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,19 +114,19 @@ def test_WeightOnlyLinear(self):
for dtype in compression_dtype:
new_model = Model()
inc_model = INCModel(new_model)
inc_model.export_compressed_model(
compressed_model = inc_model.export_compressed_model(
qweight_config_path="saved/qconfig.json",
compression_dtype=dtype,
scale_dtype=torch.float32,
use_optimum_format=False,
)
out2 = q_model(input)
torch.save(inc_model.state_dict(), "saved/tmp.pt")
torch.save(compressed_model.state_dict(), "saved/tmp.pt")
model_size2 = os.path.getsize("saved/tmp.pt") / 1024
print("WeightOnlyLinear Model size:{:.3f}M".format(model_size2))
self.assertTrue(isinstance(inc_model.model.fc1, WeightOnlyLinear))
self.assertTrue(inc_model.model.fc1.qweight.dtype == dtype)
self.assertTrue(inc_model.model.fc1.scales.dtype == torch.float32)
self.assertTrue(isinstance(compressed_model.fc1, WeightOnlyLinear))
self.assertTrue(compressed_model.fc1.qweight.dtype == dtype)
self.assertTrue(compressed_model.fc1.scales.dtype == torch.float32)
self.assertTrue(model_size1 / model_size2 > 2)
self.assertTrue(torch.all(torch.isclose(out1, out2, atol=5e-1)))

Expand All @@ -135,35 +135,35 @@ def test_WeightOnlyLinear(self):
for dim in compress_dims:
new_model = Model()
inc_model = INCModel(new_model)
inc_model.export_compressed_model(
compressed_model = inc_model.export_compressed_model(
qweight_config_path="saved/qconfig.json",
compression_dim=dim,
use_optimum_format=False,
)
out2 = q_model(input)
torch.save(inc_model.state_dict(), "saved/tmp.pt")
torch.save(compressed_model.state_dict(), "saved/tmp.pt")
model_size2 = os.path.getsize("saved/tmp.pt") / 1024
print("WeightOnlyLinear Model size:{:.3f}M".format(model_size2))
self.assertTrue(isinstance(inc_model.model.fc1, WeightOnlyLinear))
self.assertTrue(isinstance(compressed_model.fc1, WeightOnlyLinear))
if dim == 1:
self.assertTrue(inc_model.model.fc1.qweight.shape[0] == inc_model.model.fc1.out_features)
self.assertTrue(compressed_model.fc1.qweight.shape[1] != compressed_model.fc1.in_features)
else:
self.assertTrue(inc_model.model.fc1.qweight.shape[1] == inc_model.model.fc1.in_features)
self.assertTrue(compressed_model.fc1.qweight.shape[0] != compressed_model.fc1.out_features)
self.assertTrue(model_size1 / model_size2 > 2)
self.assertTrue(torch.all(torch.isclose(out1, out2, atol=5e-1)))

# test half dtype
new_model = Model()
inc_model = INCModel(new_model)
inc_model.export_compressed_model(
compressed_model = inc_model.export_compressed_model(
qweight_config_path="saved/qconfig.json",
)
out2 = q_model(input)
torch.save(inc_model.state_dict(), "saved/tmp.pt")
torch.save(compressed_model.state_dict(), "saved/tmp.pt")
model_size2 = os.path.getsize("saved/tmp.pt") / 1024
print("WeightOnlyLinear Model size:{:.3f}M".format(model_size2))
self.assertTrue(isinstance(inc_model.model.fc1, WeightOnlyLinear))
self.assertTrue(inc_model.model.fc1.scales.dtype == torch.float16)
self.assertTrue(isinstance(compressed_model.fc1, WeightOnlyLinear))
self.assertTrue(compressed_model.fc1.scales.dtype == torch.float16)
self.assertTrue(model_size1 / model_size2 > 2)
self.assertTrue(torch.all(torch.isclose(out1, out2, atol=5e-1)))

Expand Down