[tests] remove tests from libraries with deprecated support (flax, tensorflow_text, ...) (#39051)

* rm tf/flax tests

* more flax deletions

* revert fixture change

* reverted test that should not be deleted; rm tf/flax test

* revert

* fix a few add-model-like tests

* fix add-model-like checkpoint source

* a few more

* test_get_model_files_only_pt fix

* fix test_retrieve_info_for_model_with_xxx

* fix test_retrieve_model_classes

* relative paths are the devil

* add todo

tests/models/tapas/test_tokenization_tapas.py

@@ -33,7 +33,6 @@ from transformers.models.tapas.tokenization_tapas import (
 )
 from transformers.testing_utils import (
     require_pandas,
-    require_tensorflow_probability,
     require_tokenizers,
     require_torch,
     slow,
@@ -140,41 +139,6 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
         output_text = "unwanted, running"
         return input_text, output_text
 
-    @require_tensorflow_probability
-    @slow
-    def test_tf_encode_plus_sent_to_model(self):
-        from transformers import TF_MODEL_MAPPING, TOKENIZER_MAPPING
-
-        MODEL_TOKENIZER_MAPPING = merge_model_tokenizer_mappings(TF_MODEL_MAPPING, TOKENIZER_MAPPING)
-
-        tokenizers = self.get_tokenizers(do_lower_case=False)
-        for tokenizer in tokenizers:
-            with self.subTest(f"{tokenizer.__class__.__name__}"):
-                if tokenizer.__class__ not in MODEL_TOKENIZER_MAPPING:
-                    self.skipTest(f"{tokenizer.__class__} is not in the MODEL_TOKENIZER_MAPPING")
-
-                config_class, model_class = MODEL_TOKENIZER_MAPPING[tokenizer.__class__]
-                config = config_class()
-
-                if config.is_encoder_decoder or config.pad_token_id is None:
-                    self.skipTest(reason="Model is an encoder-decoder or does not have a pad token id set")
-
-                model = model_class(config)
-
-                # Make sure the model contains at least the full vocabulary size in its embedding matrix
-                self.assertGreaterEqual(model.config.vocab_size, len(tokenizer))
-
-                # Build sequence
-                first_ten_tokens = list(tokenizer.get_vocab().keys())[:10]
-                sequence = " ".join(first_ten_tokens)
-                table = self.get_table(tokenizer, length=0)
-                encoded_sequence = tokenizer.encode_plus(table, sequence, return_tensors="tf")
-                batch_encoded_sequence = tokenizer.batch_encode_plus(table, [sequence, sequence], return_tensors="tf")
-
-                # This should not fail
-                model(encoded_sequence)
-                model(batch_encoded_sequence)
-
     def test_rust_and_python_full_tokenizers(self):
         if not self.test_rust_tokenizer:
             self.skipTest(reason="test_rust_tokenizer is set to False")

tests/models/vision_text_dual_encoder/test_modeling_vision_text_dual_encoder.py

@@ -161,10 +161,6 @@ class VisionTextDualEncoderMixin:
             (text_config.num_attention_heads, input_ids.shape[-1], input_ids.shape[-1]),
         )
 
-    def assert_almost_equals(self, a: np.ndarray, b: np.ndarray, tol: float):
-        diff = np.abs(a - b).max()
-        self.assertLessEqual(diff, tol, f"Difference between torch and flax is {diff} (>= {tol}).")
-
     def test_vision_text_dual_encoder_model(self):
         inputs_dict = self.prepare_config_and_inputs()
         self.check_vision_text_dual_encoder_model(**inputs_dict)

tests/models/wav2vec2/test_modeling_wav2vec2.py

@@ -813,12 +813,6 @@ class Wav2Vec2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
             # (Even with this call, there are still memory leak by ~0.04MB)
             self.clear_torch_jit_class_registry()
 
-    @unittest.skip(
-        "Need to investigate why config.do_stable_layer_norm is set to False here when it doesn't seem to be supported"
-    )
-    def test_flax_from_pt_safetensors(self):
-        return
-
 
 @require_torch
 class Wav2Vec2RobustModelTest(ModelTesterMixin, unittest.TestCase):

tests/models/whisper/test_tokenization_whisper.py

@@ -18,7 +18,7 @@ import numpy as np
 
 from transformers.models.whisper import WhisperTokenizer, WhisperTokenizerFast
 from transformers.models.whisper.tokenization_whisper import _combine_tokens_into_words, _find_longest_common_sequence
-from transformers.testing_utils import require_flax, require_torch, slow
+from transformers.testing_utils import require_torch, slow
 
 from ...test_tokenization_common import TokenizerTesterMixin
 
@@ -588,15 +588,6 @@ class SpeechToTextTokenizerMultilinguialTest(unittest.TestCase):
         self.assertListEqual(WhisperTokenizer._convert_to_list(np_array), test_list)
         self.assertListEqual(WhisperTokenizerFast._convert_to_list(np_array), test_list)
 
-    @require_flax
-    def test_convert_to_list_jax(self):
-        import jax.numpy as jnp
-
-        test_list = [[1, 2, 3], [4, 5, 6]]
-        jax_array = jnp.array(test_list)
-        self.assertListEqual(WhisperTokenizer._convert_to_list(jax_array), test_list)
-        self.assertListEqual(WhisperTokenizerFast._convert_to_list(jax_array), test_list)
-
     @require_torch
     def test_convert_to_list_pt(self):
         import torch