chore: fix typos in the tests directory

tests/deepspeed/test_deepspeed.py

@@ -628,7 +628,7 @@ def model_init():
                 with CaptureStd() as cs:
                     trainer.hyperparameter_search(direction="maximize", n_trials=n_trials)
             self.assertIn("DeepSpeed info", cl.out, "expected DeepSpeed logger output but got none")
-            self.assertIn(f"Trial {n_trials-1} finished with value", cs.err, "expected hyperparameter_search output")
+            self.assertIn(f"Trial {n_trials - 1} finished with value", cs.err, "expected hyperparameter_search output")
             self.assertIn("Best is trial", cs.err, "expected hyperparameter_search output")
 
     # --- These tests need to run on both zero stages --- #

tests/models/bart/test_modeling_tf_bart.py

@@ -87,7 +87,7 @@ def prepare_config_and_inputs_for_common(self):
             clip_value_min=self.eos_token_id + 1,
             clip_value_max=self.vocab_size + 1,
         )
-        # Explicity add "end of sequence" to the inputs
+        # Explicitly add "end of sequence" to the inputs
         eos_tensor = tf.expand_dims(tf.constant([self.eos_token_id] * self.batch_size), 1)
         input_ids = tf.concat([input_ids, eos_tensor], axis=1)
 
@@ -225,7 +225,7 @@ def test_decoder_model_past_large_inputs(self):
         self.model_tester.check_decoder_model_past_large_inputs(*config_and_inputs)
 
     # TODO (Joao): fix me
-    @unittest.skip("Onnx compliancy broke with TF 2.10")
+    @unittest.skip("Onnx compliance broke with TF 2.10")
     def test_onnx_compliancy(self):
         pass
 

tests/models/bert/test_modeling_tf_bert.py

@@ -735,7 +735,7 @@ def test_custom_load_tf_weights(self):
             self.assertTrue(layer.split("_")[0] in ["dropout", "classifier"])
 
     # TODO (Joao): fix me
-    @unittest.skip("Onnx compliancy broke with TF 2.10")
+    @unittest.skip("Onnx compliance broke with TF 2.10")
     def test_onnx_compliancy(self):
         pass
 

tests/models/blip/test_modeling_blip.py

@@ -474,7 +474,7 @@ def test_retain_grad_hidden_states_attentions(self):
     def test_model_get_set_embeddings(self):
         pass
 
-    # override as the `logit_scale` parameter initilization is different for Blip
+    # override as the `logit_scale` parameter initialization is different for Blip
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 
@@ -483,7 +483,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         self.assertAlmostEqual(
                             param.data.item(),
@@ -988,7 +988,7 @@ def test_training_gradient_checkpointing_use_reentrant(self):
     def test_training_gradient_checkpointing_use_reentrant_false(self):
         pass
 
-    # override as the `logit_scale` parameter initilization is different for Blip
+    # override as the `logit_scale` parameter initialization is different for Blip
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 
@@ -997,7 +997,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         self.assertAlmostEqual(
                             param.data.item(),
@@ -1206,7 +1206,7 @@ def test_training_gradient_checkpointing(self):
             loss = model(**inputs).loss
             loss.backward()
 
-    # override as the `logit_scale` parameter initilization is different for Blip
+    # override as the `logit_scale` parameter initialization is different for Blip
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 
@@ -1215,7 +1215,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         self.assertAlmostEqual(
                             param.data.item(),

tests/models/blip_2/test_modeling_blip_2.py

@@ -521,7 +521,7 @@ def test_save_load_fast_init_to_base(self):
     def test_sdpa_can_dispatch_composite_models(self):
         """
         Tests if composite models dispatch correctly on SDPA/eager when requested so when loading the model.
-        This tests only by looking at layer names, as usually SDPA layers are calles "SDPAAttention".
+        This tests only by looking at layer names, as usually SDPA layers are called "SDPAAttention".
         In contrast to the above test, this one checks if the "config._attn_implamentation" is a dict after the model
         is loaded, because we manually replicate requested attn implementation on each sub-config when loading.
         See https://github.com/huggingface/transformers/pull/32238 for more info
@@ -970,7 +970,7 @@ def test_cpu_offload(self):
     def test_sdpa_can_dispatch_composite_models(self):
         """
         Tests if composite models dispatch correctly on SDPA/eager when requested so when loading the model.
-        This tests only by looking at layer names, as usually SDPA layers are calles "SDPAAttention".
+        This tests only by looking at layer names, as usually SDPA layers are called "SDPAAttention".
         In contrast to the above test, this one checks if the "config._attn_implamentation" is a dict after the model
         is loaded, because we manually replicate requested attn implementation on each sub-config when loading.
         See https://github.com/huggingface/transformers/pull/32238 for more info
@@ -1647,7 +1647,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         self.assertAlmostEqual(
                             param.data.item(),

tests/models/bloom/test_tokenization_bloom.py

@@ -135,7 +135,7 @@ def test_encodings_from_xnli_dataset(self):
     @require_jinja
     def test_tokenization_for_chat(self):
         tokenizer = self.get_rust_tokenizer()
-        tokenizer.chat_template = "{% for message in messages %}" "{{ message.content }}{{ eos_token }}" "{% endfor %}"
+        tokenizer.chat_template = "{% for message in messages %}{{ message.content }}{{ eos_token }}{% endfor %}"
         test_chats = [
             [{"role": "system", "content": "You are a helpful chatbot."}, {"role": "user", "content": "Hello!"}],
             [

tests/models/canine/test_tokenization_canine.py

@@ -64,7 +64,7 @@ def test_prepare_batch_integration(self):
     @require_torch
     def test_encoding_keys(self):
         tokenizer = self.canine_tokenizer
-        src_text = ["Once there was a man.", "He wrote a test in HuggingFace Tranformers."]
+        src_text = ["Once there was a man.", "He wrote a test in HuggingFace Transformers."]
         batch = tokenizer(src_text, padding=True, return_tensors="pt")
         # check if input_ids, attention_mask and token_type_ids are returned
         self.assertIn("input_ids", batch)

tests/models/chinese_clip/test_modeling_chinese_clip.py

@@ -596,7 +596,7 @@ def test_retain_grad_hidden_states_attentions(self):
     def test_model_get_set_embeddings(self):
         pass
 
-    # override as the `logit_scale` parameter initilization is different for CHINESE_CLIP
+    # override as the `logit_scale` parameter initialization is different for CHINESE_CLIP
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 
@@ -608,7 +608,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         self.assertAlmostEqual(
                             param.data.item(),

tests/models/clap/test_modeling_clap.py

@@ -543,7 +543,7 @@ def test_retain_grad_hidden_states_attentions(self):
     def test_model_get_set_embeddings(self):
         pass
 
-    # override as the `logit_scale` parameter initilization is different for CLAP
+    # override as the `logit_scale` parameter initialization is different for CLAP
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 
@@ -552,7 +552,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         self.assertAlmostEqual(
                             param.data.item(),

tests/models/clip/test_modeling_clip.py

@@ -761,7 +761,7 @@ def test_retain_grad_hidden_states_attentions(self):
     def test_model_get_set_embeddings(self):
         pass
 
-    # override as the `logit_scale` parameter initilization is different for CLIP
+    # override as the `logit_scale` parameter initialization is different for CLIP
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 
@@ -770,7 +770,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         self.assertAlmostEqual(
                             param.data.item(),

tests/models/clipseg/test_modeling_clipseg.py

@@ -519,7 +519,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if "logit_scale" in name:
                         self.assertAlmostEqual(
                             param.data.item(),

tests/models/clvp/test_modeling_clvp.py

@@ -500,7 +500,7 @@ def test_inputs_embeds(self):
     def test_model_get_set_embeddings(self):
         pass
 
-    # override as the `logit_scale` parameter initilization is different for Clvp
+    # override as the `logit_scale` parameter initialization is different for Clvp
     def test_initialization(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 
@@ -509,7 +509,7 @@ def test_initialization(self):
             model = model_class(config=configs_no_init)
             for name, param in model.named_parameters():
                 if param.requires_grad:
-                    # check if `logit_scale` is initilized as per the original implementation
+                    # check if `logit_scale` is initialized as per the original implementation
                     if name == "logit_scale":
                         expected_value = np.log(1 / 0.07)
                         returned_value = param.data.item()

tests/models/conditional_detr/test_modeling_conditional_detr.py

@@ -465,13 +465,13 @@ def test_different_timm_backbone(self):
                     self.model_tester.num_labels,
                 )
                 self.assertEqual(outputs.logits.shape, expected_shape)
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             elif model_class.__name__ == "ConditionalDetrForSegmentation":
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.conditional_detr.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             else:
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.backbone.conv_encoder.intermediate_channel_sizes), 3)
 
             self.assertTrue(outputs)
@@ -501,13 +501,13 @@ def test_hf_backbone(self):
                     self.model_tester.num_labels,
                 )
                 self.assertEqual(outputs.logits.shape, expected_shape)
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             elif model_class.__name__ == "ConditionalDetrForSegmentation":
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.conditional_detr.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             else:
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.backbone.conv_encoder.intermediate_channel_sizes), 3)
 
             self.assertTrue(outputs)

tests/models/cpmant/test_modeling_cpmant.py

@@ -161,7 +161,7 @@ def test_inputs_embeds(self):
     def test_retain_grad_hidden_states_attentions(self):
         unittest.skip(
             "CPMAnt doesn't support retain grad in hidden_states or attentions, because prompt management will peel off the output.hidden_states from graph.\
-                 So is attentions. We strongly recommand you use loss to tune model."
+                 So is attentions. We strongly recommend you use loss to tune model."
         )(self.test_retain_grad_hidden_states_attentions)
 
     def test_cpmant_model(self):

tests/models/deformable_detr/test_modeling_deformable_detr.py

@@ -542,10 +542,10 @@ def test_different_timm_backbone(self):
                     self.model_tester.num_labels,
                 )
                 self.assertEqual(outputs.logits.shape, expected_shape)
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.model.backbone.conv_encoder.intermediate_channel_sizes), 4)
             else:
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.backbone.conv_encoder.intermediate_channel_sizes), 4)
 
             self.assertTrue(outputs)
@@ -574,10 +574,10 @@ def test_hf_backbone(self):
                     self.model_tester.num_labels,
                 )
                 self.assertEqual(outputs.logits.shape, expected_shape)
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.model.backbone.conv_encoder.intermediate_channel_sizes), 4)
             else:
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.backbone.conv_encoder.intermediate_channel_sizes), 4)
 
             self.assertTrue(outputs)

tests/models/depth_anything/test_modeling_depth_anything.py

@@ -214,7 +214,7 @@ def _validate_backbone_init():
                 model.to(torch_device)
                 model.eval()
 
-                # Confirm out_indices propogated to backbone
+                # Confirm out_indices propagated to backbone
                 self.assertEqual(len(model.backbone.out_indices), 2)
 
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()

tests/models/depth_pro/test_modeling_depth_pro.py

@@ -323,7 +323,7 @@ def test_initialization(self):
                             msg=f"Parameter {name} of model {model_class} seems not properly initialized",
                         )
 
-    # this started when switched from normal initialization to kaiming_normal intialization
+    # this started when switched from normal initialization to kaiming_normal initialization
     # maybe because the magnitude of offset values from ViT-encoders increases when followed by many convolution layers
     def test_batching_equivalence(self, atol=1e-4, rtol=1e-4):
         super().test_batching_equivalence(atol=atol, rtol=rtol)

tests/models/detr/test_modeling_detr.py

@@ -465,13 +465,13 @@ def test_different_timm_backbone(self):
                     self.model_tester.num_labels + 1,
                 )
                 self.assertEqual(outputs.logits.shape, expected_shape)
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             elif model_class.__name__ == "DetrForSegmentation":
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.detr.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             else:
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.backbone.conv_encoder.intermediate_channel_sizes), 3)
 
             self.assertTrue(outputs)
@@ -500,13 +500,13 @@ def test_hf_backbone(self):
                     self.model_tester.num_labels + 1,
                 )
                 self.assertEqual(outputs.logits.shape, expected_shape)
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             elif model_class.__name__ == "DetrForSegmentation":
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.detr.model.backbone.conv_encoder.intermediate_channel_sizes), 3)
             else:
-                # Confirm out_indices was propogated to backbone
+                # Confirm out_indices was propagated to backbone
                 self.assertEqual(len(model.backbone.conv_encoder.intermediate_channel_sizes), 3)
 
             self.assertTrue(outputs)