Fix HF tests with Pin (mosaicml#3248)

* fix tests

* pin hub

* ignore warnings

composer/models/huggingface.py

@@ -71,8 +71,8 @@ class HuggingFaceModel(ComposerModel):
         import transformers
         from composer.models import HuggingFaceModel
 
-        hf_model = transformers.AutoModelForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=2)
-        hf_tokenizer = transformers.AutoTokenizer.from_pretrained('bert-base-uncased')
+        hf_model = transformers.AutoModelForSequenceClassification.from_pretrained('google-bert/bert-base-uncased', num_labels=2)
+        hf_tokenizer = transformers.AutoTokenizer.from_pretrained('google-bert/bert-base-uncased')
         model = HuggingFaceModel(hf_model, hf_tokenizer)
     """
 

docs/source/composer_model.rst

@@ -210,7 +210,7 @@ and make it compatible with our trainer.
 
     # huggingface model
     model = AutoModelForSequenceClassification.from_pretrained(
-                            'bert-base-uncased',
+                            'google-bert/bert-base-uncased',
                              num_labels=2)
 
     # list of torchmetrics

examples/finetune_huggingface.ipynb

@@ -77,8 +77,8 @@
     "import transformers\n",
     "\n",
     "# Create a BERT sequence classification model using Hugging Face transformers\n",
-    "model = transformers.AutoModelForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=2)\n",
-    "tokenizer = transformers.AutoTokenizer.from_pretrained('bert-base-uncased') "
+    "model = transformers.AutoModelForSequenceClassification.from_pretrained('google-bert/bert-base-uncased', num_labels=2)\n",
+    "tokenizer = transformers.AutoTokenizer.from_pretrained('google-bert/bert-base-uncased') "
    ]
   },
   {

pyproject.toml

@@ -159,7 +159,9 @@ filterwarnings = [
     # Ignore autograd kernel warning inside DeepSpeed
     '''ignore:.*an autograd kernel was not registered to the Autograd key.*:UserWarning''',
     # Ignore save_state_dict / load_state_dict deprecation warnings
-    '''ignore:'.*_state_dict' is deprecated and will be removed in future versions.*:UserWarning'''
+    '''ignore:'.*_state_dict' is deprecated and will be removed in future versions.*:UserWarning''',
+    # Ignore HF deprecation which affects their own libraries
+    '''ignore:'.*`resume_download` is deprecated and will be removed in version.*:FutureWarning'''
 ]
 
 # Coverage

setup.py

@@ -181,6 +181,7 @@ def package_files(prefix: str, directory: str, extension: str):
 extra_deps['nlp'] = [
     'transformers>=4.11,!=4.34.0,<4.41',
     'datasets>=2.4,<3',
+    'huggingface-hub>=0.21.2,<0.23',
 ]
 
 extra_deps['peft'] = [

tests/fixtures/fixtures.py

@@ -146,7 +146,7 @@ def _session_tiny_bert_model(_session_tiny_bert_config):  # type: ignore
 def tiny_bert_tokenizer_helper():
     transformers = pytest.importorskip('transformers')
 
-    return transformers.AutoTokenizer.from_pretrained('bert-base-uncased')
+    return transformers.AutoTokenizer.from_pretrained('google-bert/bert-base-uncased')
 
 
 @pytest.fixture(scope='session')
@@ -162,7 +162,7 @@ def tiny_bert_config_helper():
         'num_hidden_layers': 2,
         'intermediate_size': 512,
     }
-    return transformers.AutoConfig.from_pretrained('bert-base-uncased', **tiny_overrides)
+    return transformers.AutoConfig.from_pretrained('google-bert/bert-base-uncased', **tiny_overrides)
 
 
 @pytest.fixture(scope='session')
@@ -302,7 +302,7 @@ def tiny_t5_config_helper():
     transformers = pytest.importorskip('transformers')
 
     tiny_overrides = {'d_ff': 128, 'd_model': 64, 'num_layers': 2, 'num_decoder_layers': 2, 'num_heads': 2}
-    return transformers.AutoConfig.from_pretrained('t5-small', **tiny_overrides)
+    return transformers.AutoConfig.from_pretrained('google-t5/t5-small', **tiny_overrides)
 
 
 @pytest.fixture(scope='session')
@@ -313,7 +313,7 @@ def _session_tiny_t5_config():  # type: ignore
 def tiny_t5_tokenizer_helper():
     transformers = pytest.importorskip('transformers')
 
-    hf_tokenizer = transformers.AutoTokenizer.from_pretrained('t5-small', model_max_length=512)
+    hf_tokenizer = transformers.AutoTokenizer.from_pretrained('google-t5/t5-small', model_max_length=512)
     return hf_tokenizer