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

use torch.testing.assertclose instead to get more details about error in cis (#35659)

Browse Source 
* use torch.testing.assertclose instead to get more details about error in cis

* fix

* style

* test_all

* revert for I bert

* fixes and updates

* more image processing fixes

* more image processors

* fix mamba and co

* style

* less strick

* ok I won't be strict

* skip and be done

* up
This commit is contained in:
Arthur
2025-01-24 16:55:28 +01:00
committed by GitHub
parent 72d1a4cd53
commit b912f5ee43
255 changed files with 1048 additions and 969 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/utils/test_activations.py
View File

@@ -29,7 +29,7 @@ class TestActivations(unittest.TestCase):
def test_gelu_versions(self):
x = torch.tensor([-100, -1, -0.1, 0, 0.1, 1.0, 100])
torch_builtin = get_activation("gelu")
self.assertTrue(torch.allclose(gelu_python(x), torch_builtin(x)))
torch.testing.assert_close(gelu_python(x), torch_builtin(x))
self.assertFalse(torch.allclose(gelu_python(x), gelu_new(x)))
def test_gelu_10(self):
@@ -43,7 +43,7 @@ class TestActivations(unittest.TestCase):
clipped_mask = torch.where(y_gelu_10 < 10.0, 1, 0)
self.assertTrue(torch.max(y_gelu_10).item() == 10.0)
self.assertTrue(torch.allclose(y_gelu * clipped_mask, y_gelu_10 * clipped_mask))
torch.testing.assert_close(y_gelu * clipped_mask, y_gelu_10 * clipped_mask)
def test_get_activation(self):
get_activation("gelu")

42
tests/utils/test_modeling_utils.py
View File

@@ -742,14 +742,14 @@ class ModelUtilsTest(TestCasePlus):
# Finally, check the model can be reloaded
new_model = BertModel.from_pretrained(tmp_dir)
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
def test_checkpoint_sharding_from_hub(self):
model = BertModel.from_pretrained("hf-internal-testing/tiny-random-bert-sharded")
# the model above is the same as the model below, just a sharded version.
ref_model = BertModel.from_pretrained("hf-internal-testing/tiny-random-bert")
for p1, p2 in zip(model.parameters(), ref_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
def test_checkpoint_variant_local_bin(self):
model = BertModel.from_pretrained("hf-internal-testing/tiny-random-bert")
@@ -769,7 +769,7 @@ class ModelUtilsTest(TestCasePlus):
new_model = BertModel.from_pretrained(tmp_dir, variant="v2")
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
def test_checkpoint_variant_local_sharded_bin(self):
model = BertModel.from_pretrained("hf-internal-testing/tiny-random-bert")
@@ -793,7 +793,7 @@ class ModelUtilsTest(TestCasePlus):
new_model = BertModel.from_pretrained(tmp_dir, variant="v2")
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
@require_safetensors
def test_checkpoint_variant_local_safe(self):
@@ -814,7 +814,7 @@ class ModelUtilsTest(TestCasePlus):
new_model = BertModel.from_pretrained(tmp_dir, variant="v2")
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
@require_safetensors
def test_checkpoint_variant_local_sharded_safe(self):
@@ -839,7 +839,7 @@ class ModelUtilsTest(TestCasePlus):
new_model = BertModel.from_pretrained(tmp_dir, variant="v2")
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
def test_checkpoint_loading_only_safetensors_available(self):
# Test that the loading behaviour is as expected when only safetensor checkpoints are available
@@ -872,7 +872,7 @@ class ModelUtilsTest(TestCasePlus):
new_model = BertModel.from_pretrained(tmp_dir)
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
def test_checkpoint_loading_only_pytorch_bin_available(self):
# Test that the loading behaviour is as expected when only pytorch checkpoints are available
@@ -905,7 +905,7 @@ class ModelUtilsTest(TestCasePlus):
new_model = BertModel.from_pretrained(tmp_dir)
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
def test_checkpoint_variant_hub(self):
with tempfile.TemporaryDirectory() as tmp_dir:
@@ -1068,7 +1068,7 @@ class ModelUtilsTest(TestCasePlus):
)
outputs2 = new_model_with_offload(inputs)
self.assertTrue(torch.allclose(outputs1.logits.cpu(), outputs2.logits.cpu()))
torch.testing.assert_close(outputs1.logits.cpu(), outputs2.logits.cpu())
# With state dict temp offload
new_model_with_offload = AutoModelForCausalLM.from_pretrained(
@@ -1078,7 +1078,7 @@ class ModelUtilsTest(TestCasePlus):
offload_state_dict=True,
)
outputs2 = new_model_with_offload(inputs)
self.assertTrue(torch.allclose(outputs1.logits.cpu(), outputs2.logits.cpu()))
torch.testing.assert_close(outputs1.logits.cpu(), outputs2.logits.cpu())
@require_accelerate
@mark.accelerate_tests
@@ -1108,7 +1108,7 @@ class ModelUtilsTest(TestCasePlus):
tmp_dir, device_map=device_map, offload_folder=offload_folder
)
outputs2 = base_model_with_offload(inputs)
self.assertTrue(torch.allclose(outputs1[0].cpu(), outputs2[0].cpu()))
torch.testing.assert_close(outputs1[0].cpu(), outputs2[0].cpu())
# With state dict temp offload
new_model_with_offload = AutoModel.from_pretrained(
@@ -1118,7 +1118,7 @@ class ModelUtilsTest(TestCasePlus):
offload_state_dict=True,
)
outputs2 = new_model_with_offload(inputs)
self.assertTrue(torch.allclose(outputs1[0].cpu(), outputs2[0].cpu()))
torch.testing.assert_close(outputs1[0].cpu(), outputs2[0].cpu())
@slow
@require_torch
@@ -1169,7 +1169,7 @@ class ModelUtilsTest(TestCasePlus):
saved_model = AutoModelForCausalLM.from_pretrained(tmp_dir, device_map="cpu")
saved_model_output = saved_model(inputs)[0]
self.assertTrue(torch.allclose(output, saved_model_output))
torch.testing.assert_close(output, saved_model_output)
@require_accelerate
@mark.accelerate_tests
@@ -1205,8 +1205,8 @@ class ModelUtilsTest(TestCasePlus):
saved_model = AutoModelForCausalLM.from_pretrained(tmp_dir, device_map=device_map)
postsaved_output = saved_model(inputs)[0]
self.assertTrue(torch.allclose(output, presaved_output, atol=1e-4))
self.assertTrue(torch.allclose(presaved_output, postsaved_output))
torch.testing.assert_close(output, presaved_output, rtol=1e-4, atol=1e-4)
torch.testing.assert_close(presaved_output, postsaved_output)
@require_safetensors
def test_use_safetensors(self):
@@ -1278,7 +1278,7 @@ class ModelUtilsTest(TestCasePlus):
# Check models are equal
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
@require_safetensors
def test_safetensors_load_from_hub(self):
@@ -1287,7 +1287,7 @@ class ModelUtilsTest(TestCasePlus):
# Check models are equal
for p1, p2 in zip(safetensors_model.parameters(), pytorch_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
@require_safetensors
def test_safetensors_save_and_load_sharded(self):
@@ -1305,7 +1305,7 @@ class ModelUtilsTest(TestCasePlus):
# Check models are equal
for p1, p2 in zip(model.parameters(), new_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
@require_safetensors
def test_safetensors_load_from_hub_sharded(self):
@@ -1314,7 +1314,7 @@ class ModelUtilsTest(TestCasePlus):
# Check models are equal
for p1, p2 in zip(safetensors_model.parameters(), pytorch_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
def test_base_model_to_head_model_load(self):
base_model = BaseModel(PretrainedConfig())
@@ -1324,7 +1324,7 @@ class ModelUtilsTest(TestCasePlus):
# Can load a base model in a model with head
model = ModelWithHead.from_pretrained(tmp_dir)
for p1, p2 in zip(model.base.parameters(), base_model.parameters()):
self.assertTrue(torch.allclose(p1, p2))
torch.testing.assert_close(p1, p2)
# It doesn't work if the state dict has a mix of keys of the head and base without prefix though.
base_state_dict = base_model.state_dict()
@@ -1615,7 +1615,7 @@ class ModelUtilsTest(TestCasePlus):
with torch.no_grad():
outputs = model(input_ids)
outputs_from_saved = new_model(input_ids)
self.assertTrue(torch.allclose(outputs_from_saved["logits"], outputs["logits"]))
torch.testing.assert_close(outputs_from_saved["logits"], outputs["logits"])
def test_warning_for_beta_gamma_parameters(self):
class TestGammaBetaNorm(torch.nn.Module):