SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Add 'with torch.no_grad()' to BEiT integration test forward passes (#14961)

Browse Source 
* Add 'with torch.no_grad()' to BEiT integration test forward pass

* Fix inconsistent use of tabs and spaces in indentation
This commit is contained in:
Tavin Turner
2022-01-31 13:12:10 -07:00
committed by GitHub
parent 09f9d07271
commit d984b10335
1 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
tests/test_modeling_beit.py
View File

@@ -435,7 +435,8 @@ class BeitModelIntegrationTest(unittest.TestCase):
bool_masked_pos = torch.ones((1, 196), dtype=torch.bool).to(torch_device) bool_masked_pos = torch.ones((1, 196), dtype=torch.bool).to(torch_device)
# forward pass # forward pass
outputs = model(pixel_values=pixel_values, bool_masked_pos=bool_masked_pos) with torch.no_grad():
outputs = model(pixel_values=pixel_values, bool_masked_pos=bool_masked_pos)
logits = outputs.logits logits = outputs.logits
# verify the logits # verify the logits
@@ -457,7 +458,8 @@ class BeitModelIntegrationTest(unittest.TestCase):
inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device) inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device)
# forward pass # forward pass
outputs = model(**inputs) with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits logits = outputs.logits
# verify the logits # verify the logits
@@ -482,7 +484,8 @@ class BeitModelIntegrationTest(unittest.TestCase):
inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device) inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device)
# forward pass # forward pass
outputs = model(**inputs) with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits logits = outputs.logits
# verify the logits # verify the logits
@@ -508,7 +511,8 @@ class BeitModelIntegrationTest(unittest.TestCase):
inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device) inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device)
# forward pass # forward pass
outputs = model(**inputs) with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits logits = outputs.logits
# verify the logits # verify the logits