Merge branch 'huggingface:main' into fix_num_assistant_tokens

src/transformers/generation/utils.py

@@ -135,7 +135,7 @@ class GenerateDecoderOnlyOutput(ModelOutput):
 @dataclass
 class GenerateEncoderDecoderOutput(ModelOutput):
     """
-    Outputs of encoder-decider generation models, when using non-beam methods.
+    Outputs of encoder-decoder generation models, when using non-beam methods.
 
     Args:
         sequences (`torch.LongTensor` of shape `(batch_size, sequence_length)`):

src/transformers/modeling_utils.py

@@ -4192,7 +4192,7 @@ def warn_if_padding_and_no_attention_mask(self, input_ids, attention_mask):
 
     @property
     def _is_quantized_training_enabled(self):
-        logger.warning(
+        warnings.warn(
             "`_is_quantized_training_enabled` is going to be deprecated in transformers 4.39.0. Please use `model.hf_quantizer.is_trainable` instead",
             FutureWarning,
         )

src/transformers/models/deta/modeling_deta.py

@@ -627,7 +627,8 @@ def __init__(self, config: DetaConfig, num_heads: int, n_points: int):
 
     def _reset_parameters(self):
         nn.init.constant_(self.sampling_offsets.weight.data, 0.0)
-        thetas = torch.arange(self.n_heads, dtype=torch.int64).float() * (2.0 * math.pi / self.n_heads)
+        default_dtype = torch.get_default_dtype()
+        thetas = torch.arange(self.n_heads, dtype=torch.int64).to(default_dtype) * (2.0 * math.pi / self.n_heads)
         grid_init = torch.stack([thetas.cos(), thetas.sin()], -1)
         grid_init = (
             (grid_init / grid_init.abs().max(-1, keepdim=True)[0])

src/transformers/quantizers/quantizer_aqlm.py

@@ -77,7 +77,6 @@ def _process_model_before_weight_loading(
         model.config.quantization_config = self.quantization_config
 
     def _process_model_after_weight_loading(self, model: "PreTrainedModel", **kwargs):
-        model._is_quantized_training_enabled = False
         return model
 
     @property

tests/models/bert_generation/test_modeling_bert_generation.py

@@ -305,6 +305,12 @@ def test_model_from_pretrained(self):
         model = BertGenerationEncoder.from_pretrained("google/bert_for_seq_generation_L-24_bbc_encoder")
         self.assertIsNotNone(model)
 
+    @unittest.skip(
+        "Not currently compatible. Fails with - NotImplementedError: Cannot copy out of meta tensor; no data!"
+    )
+    def test_save_load_low_cpu_mem_usage(self):
+        pass
+
 
 @require_torch
 class BertGenerationEncoderIntegrationTest(unittest.TestCase):

tests/models/fsmt/test_modeling_fsmt.py

@@ -329,6 +329,12 @@ def test_tie_model_weights(self):
     def test_resize_embeddings_untied(self):
         pass
 
+    @unittest.skip(
+        "Not currently compatible. Fails with - NotImplementedError: Cannot copy out of meta tensor; no data!"
+    )
+    def test_save_load_low_cpu_mem_usage(self):
+        pass
+
 
 @require_torch
 class FSMTHeadTests(unittest.TestCase):

tests/models/marian/test_modeling_marian.py

@@ -372,6 +372,12 @@ def test_training_gradient_checkpointing_use_reentrant(self):
     def test_training_gradient_checkpointing_use_reentrant_false(self):
         pass
 
+    @unittest.skip(
+        "Not currently compatible. Fails with - NotImplementedError: Cannot copy out of meta tensor; no data!"
+    )
+    def test_save_load_low_cpu_mem_usage(self):
+        pass
+
 
 def assert_tensors_close(a, b, atol=1e-12, prefix=""):
     """If tensors have different shapes, different values or a and b are not both tensors, raise a nice Assertion error."""

tests/models/musicgen/test_modeling_musicgen.py

@@ -1144,6 +1144,10 @@ def test_greedy_generate_stereo_outputs(self):
 
             self.assertNotIn(config.pad_token_id, output_generate)
 
+    @unittest.skip("Fails with - TypeError: _weight_norm_interface() missing 1 required positional argument: 'dim'")
+    def test_save_load_low_cpu_mem_usage(self):
+        pass
+
 
 def get_bip_bip(bip_duration=0.125, duration=0.5, sample_rate=32000):
     """Produces a series of 'bip bip' sounds at a given frequency."""

tests/models/reformer/test_modeling_reformer.py

@@ -687,6 +687,12 @@ def _check_hidden_states_for_generate(
     def test_left_padding_compatibility(self):
         pass
 
+    @unittest.skip(
+        "Not currently compatible. Fails with - NotImplementedError: Cannot copy out of meta tensor; no data!"
+    )
+    def test_save_load_low_cpu_mem_usage(self):
+        pass
+
 
 @require_torch
 class ReformerLSHAttnModelTest(
@@ -848,6 +854,12 @@ def test_past_key_values_format(self):
     def test_left_padding_compatibility(self):
         pass
 
+    @unittest.skip(
+        "Not currently compatible. Fails with - NotImplementedError: Cannot copy out of meta tensor; no data!"
+    )
+    def test_save_load_low_cpu_mem_usage(self):
+        pass
+
 
 @require_torch
 @require_sentencepiece

tests/models/xlm_roberta_xl/test_modeling_xlm_roberta_xl.py

@@ -515,6 +515,12 @@ def test_create_position_ids_from_inputs_embeds(self):
         self.assertEqual(position_ids.shape, expected_positions.shape)
         self.assertTrue(torch.all(torch.eq(position_ids, expected_positions)))
 
+    @unittest.skip(
+        "Not currently compatible. Fails with - NotImplementedError: Cannot copy out of meta tensor; no data!"
+    )
+    def test_save_load_low_cpu_mem_usage(self):
+        pass
+
 
 @require_torch
 class XLMRobertaModelXLIntegrationTest(unittest.TestCase):