Revert workaround for TF safetensors loading (#30128)

* See if we can get tests to pass with the fixed weights

* See if we can get tests to pass with the fixed weights

* Replace the revisions now that we don't need them anymore

tests/generation/test_framework_agnostic.py

@@ -111,7 +111,7 @@ class GenerationIntegrationTestsMixin:
         article = """Justin Timberlake."""
         gpt2_tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-gpt2")
 
-        gpt2_model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2", use_safetensors=is_pt)
+        gpt2_model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         input_ids = gpt2_tokenizer(article, return_tensors=return_tensors).input_ids
         if is_pt:
             gpt2_model = gpt2_model.to(torch_device)
@@ -582,7 +582,7 @@ class GenerationIntegrationTestsMixin:
         tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         text = """Hello, my dog is cute and"""
         tokens = tokenizer(text, return_tensors=return_tensors)
-        model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2", use_safetensors=is_pt)
+        model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         if is_pt:
             model = model.to(torch_device)
             tokens = tokens.to(torch_device)
@@ -611,7 +611,7 @@ class GenerationIntegrationTestsMixin:
         tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         text = """Hello, my dog is cute and"""
         tokens = tokenizer(text, return_tensors=return_tensors)
-        model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2", use_safetensors=is_pt)
+        model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         if is_pt:
             model = model.to(torch_device)
             tokens = tokens.to(torch_device)
@@ -638,7 +638,7 @@ class GenerationIntegrationTestsMixin:
         tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         text = """Hello, my dog is cute and"""
         tokens = tokenizer(text, return_tensors=return_tensors)
-        model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2", use_safetensors=is_pt)
+        model = model_cls.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         if is_pt:
             model = model.to(torch_device)
             tokens = tokens.to(torch_device)

tests/generation/test_tf_utils.py

@@ -194,7 +194,7 @@ class TFGenerationIntegrationTests(unittest.TestCase, GenerationIntegrationTests
         tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-gpt2")
         text = """Hello, my dog is cute and"""
         tokens = tokenizer(text, return_tensors="tf")
-        model = TFAutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-gpt2", use_safetensors=False)
+        model = TFAutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-gpt2")
 
         eos_token_id = 638
         # forces the generation to happen on CPU, to avoid GPU-related quirks

tests/pipelines/test_pipelines_text_generation.py

@@ -268,7 +268,6 @@ class TextGenerationPipelineTests(unittest.TestCase):
         text_generator = TextGenerationPipeline(model=model, tokenizer=tokenizer)
         return text_generator, ["This is a test", "Another test"]
 
-    @require_torch  # See https://github.com/huggingface/transformers/issues/30117
     def test_stop_sequence_stopping_criteria(self):
         prompt = """Hello I believe in"""
         text_generator = pipeline("text-generation", model="hf-internal-testing/tiny-random-gpt2")