Make Transformers use cache files when hf.co is down (#16362)

* Make Transformers use cache files when hf.co is down

* Fix tests

* Was there a random circleCI failure?

* Isolate patches

* Style

* Comment out the failure since it doesn't fail anymore

* Better comment

tests/test_configuration_common.py

@@ -20,7 +20,7 @@ import shutil
 import sys
 import tempfile
 import unittest
-import unittest.mock
+import unittest.mock as mock
 from pathlib import Path
 
 from huggingface_hub import Repository, delete_repo, login
@@ -304,6 +304,22 @@ class ConfigTestUtils(unittest.TestCase):
                 f"pick another value for them: {', '.join(keys_with_defaults)}."
             )
 
+    def test_cached_files_are_used_when_internet_is_down(self):
+        # A mock response for an HTTP head request to emulate server down
+        response_mock = mock.Mock()
+        response_mock.status_code = 500
+        response_mock.headers = []
+        response_mock.raise_for_status.side_effect = HTTPError
+
+        # Download this model to make sure it's in the cache.
+        _ = BertConfig.from_pretrained("hf-internal-testing/tiny-random-bert")
+
+        # Under the mock environment we get a 500 error when trying to reach the model.
+        with mock.patch("transformers.utils.hub.requests.head", return_value=response_mock) as mock_head:
+            _ = BertConfig.from_pretrained("hf-internal-testing/tiny-random-bert")
+            # This check we did call the fake head request
+            mock_head.assert_called()
+
 
 class ConfigurationVersioningTest(unittest.TestCase):
     def test_local_versioning(self):

tests/test_feature_extraction_common.py

@@ -19,6 +19,7 @@ import os
 import sys
 import tempfile
 import unittest
+import unittest.mock as mock
 from pathlib import Path
 
 from huggingface_hub import Repository, delete_repo, login
@@ -116,6 +117,23 @@ class FeatureExtractionSavingTestMixin:
         self.assertIsNotNone(feat_extract)
 
 
+class FeatureExtractorUtilTester(unittest.TestCase):
+    def test_cached_files_are_used_when_internet_is_down(self):
+        # A mock response for an HTTP head request to emulate server down
+        response_mock = mock.Mock()
+        response_mock.status_code = 500
+        response_mock.headers = []
+        response_mock.raise_for_status.side_effect = HTTPError
+
+        # Download this model to make sure it's in the cache.
+        _ = Wav2Vec2FeatureExtractor.from_pretrained("hf-internal-testing/tiny-random-wav2vec2")
+        # Under the mock environment we get a 500 error when trying to reach the model.
+        with mock.patch("transformers.utils.hub.requests.head", return_value=response_mock) as mock_head:
+            _ = Wav2Vec2FeatureExtractor.from_pretrained("hf-internal-testing/tiny-random-wav2vec2")
+            # This check we did call the fake head request
+            mock_head.assert_called()
+
+
 @is_staging_test
 class FeatureExtractorPushToHubTester(unittest.TestCase):
     @classmethod

tests/test_modeling_common.py

@@ -23,6 +23,7 @@ import random
 import sys
 import tempfile
 import unittest
+import unittest.mock as mock
 import warnings
 from pathlib import Path
 from typing import Dict, List, Tuple
@@ -2272,6 +2273,22 @@ class ModelUtilsTest(TestCasePlus):
         for p1, p2 in zip(model.parameters(), new_model.parameters()):
             self.assertTrue(torch.equal(p1, p2))
 
+    def test_cached_files_are_used_when_internet_is_down(self):
+        # A mock response for an HTTP head request to emulate server down
+        response_mock = mock.Mock()
+        response_mock.status_code = 500
+        response_mock.headers = []
+        response_mock.raise_for_status.side_effect = HTTPError
+
+        # Download this model to make sure it's in the cache.
+        _ = BertModel.from_pretrained("hf-internal-testing/tiny-random-bert")
+
+        # Under the mock environment we get a 500 error when trying to reach the model.
+        with mock.patch("transformers.utils.hub.requests.head", return_value=response_mock) as mock_head:
+            _ = BertModel.from_pretrained("hf-internal-testing/tiny-random-bert")
+            # This check we did call the fake head request
+            mock_head.assert_called()
+
 
 @require_torch
 @is_staging_test

tests/test_modeling_tf_common.py

@@ -21,6 +21,7 @@ import os
 import random
 import tempfile
 import unittest
+import unittest.mock as mock
 from importlib import import_module
 from typing import List, Tuple
 
@@ -1555,6 +1556,22 @@ class UtilsFunctionsTest(unittest.TestCase):
         tf.debugging.assert_near(non_inf_output, non_inf_expected_output, rtol=1e-12)
         tf.debugging.assert_equal(non_inf_idx, non_inf_expected_idx)
 
+    def test_cached_files_are_used_when_internet_is_down(self):
+        # A mock response for an HTTP head request to emulate server down
+        response_mock = mock.Mock()
+        response_mock.status_code = 500
+        response_mock.headers = []
+        response_mock.raise_for_status.side_effect = HTTPError
+
+        # Download this model to make sure it's in the cache.
+        _ = TFBertModel.from_pretrained("hf-internal-testing/tiny-random-bert")
+
+        # Under the mock environment we get a 500 error when trying to reach the model.
+        with mock.patch("transformers.utils.hub.requests.head", return_value=response_mock) as mock_head:
+            _ = TFBertModel.from_pretrained("hf-internal-testing/tiny-random-bert")
+            # This check we did call the fake head request
+            mock_head.assert_called()
+
     # tests whether the unpack_inputs function behaves as expected
     def test_unpack_inputs(self):
         class DummyModel:

tests/test_tokenization_common.py

@@ -24,6 +24,7 @@ import shutil
 import sys
 import tempfile
 import unittest
+import unittest.mock as mock
 from collections import OrderedDict
 from itertools import takewhile
 from pathlib import Path
@@ -3742,6 +3743,24 @@ class TokenizerTesterMixin:
                     self.rust_tokenizer_class.from_pretrained(tmp_dir_2)
 
 
+class TokenizerUtilTester(unittest.TestCase):
+    def test_cached_files_are_used_when_internet_is_down(self):
+        # A mock response for an HTTP head request to emulate server down
+        response_mock = mock.Mock()
+        response_mock.status_code = 500
+        response_mock.headers = []
+        response_mock.raise_for_status.side_effect = HTTPError
+
+        # Download this model to make sure it's in the cache.
+        _ = BertTokenizer.from_pretrained("hf-internal-testing/tiny-random-bert")
+
+        # Under the mock environment we get a 500 error when trying to reach the model.
+        with mock.patch("transformers.utils.hub.requests.head", return_value=response_mock) as mock_head:
+            _ = BertTokenizer.from_pretrained("hf-internal-testing/tiny-random-bert")
+            # This check we did call the fake head request
+            mock_head.assert_called()
+
+
 @is_staging_test
 class TokenizerPushToHubTester(unittest.TestCase):
     vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]", "bla", "blou"]

tests/utils/test_offline.py

@@ -59,10 +59,10 @@ socket.socket = offline_socket
         # next emulate no network
         cmd = [sys.executable, "-c", "\n".join([load, mock, run])]
 
-        # should normally fail as it will fail to lookup the model files w/o the network
-        env["TRANSFORMERS_OFFLINE"] = "0"
-        result = subprocess.run(cmd, env=env, check=False, capture_output=True)
-        self.assertEqual(result.returncode, 1, result.stderr)
+        # Doesn't fail anymore since the model is in the cache due to other tests, so commenting this.
+        # env["TRANSFORMERS_OFFLINE"] = "0"
+        # result = subprocess.run(cmd, env=env, check=False, capture_output=True)
+        # self.assertEqual(result.returncode, 1, result.stderr)
 
         # should succeed as TRANSFORMERS_OFFLINE=1 tells it to use local files
         env["TRANSFORMERS_OFFLINE"] = "1"