Use removeprefix and removesuffix (#41240)

* Use removeprefix and removesuffix

Signed-off-by: Yuanyuan Chen <cyyever@outlook.com>

* More fixes

Signed-off-by: Yuanyuan Chen <cyyever@outlook.com>

---------

Signed-off-by: Yuanyuan Chen <cyyever@outlook.com>

Author: cyyever

src/transformers/commands/chat.py

@@ -443,8 +443,7 @@ def parse_generate_flags(self, generate_flags: list[str]) -> dict:
         # 2. b. strings should be quoted
         def is_number(s: str) -> bool:
             # handle negative numbers
-            if s.startswith("-"):
-                s = s[1:]
+            s = s.removeprefix("-")
             return s.replace(".", "", 1).isdigit()
 
         generate_flags_as_dict = {k: f'"{v}"' if not is_number(v) else v for k, v in generate_flags_as_dict.items()}

src/transformers/commands/serving.py

@@ -1066,8 +1066,7 @@ def generate_with_cache(**kwargs):
                 for result in streamer:
                     # Temporary hack for GPTOS 3: don't emit the final "<|return|>"
                     if "gptoss" in model.config.architectures[0].lower():
-                        if result.endswith("<|return|>"):
-                            result = result[: -len("<|return|>")]
+                        result = result.removesuffix("<|return|>")
                     results += result
 
                     # (related to temporary hack 2)
@@ -1325,8 +1324,7 @@ def generate_with_cache(**kwargs):
                 for result in streamer:
                     # Temporary hack for GPTOS 3: don't emit the final "<|return|>"
                     if "gptoss" in model.config.architectures[0].lower():
-                        if result.endswith("<|return|>"):
-                            result = result[: -len("<|return|>")]
+                        result = result.removesuffix("<|return|>")
                     results += result
 
                     # (related to temporary hack 2)

src/transformers/dynamic_module_utils.py

@@ -285,8 +285,7 @@ def get_class_in_module(
         `typing.Type`: The class looked for.
     """
     name = os.path.normpath(module_path)
-    if name.endswith(".py"):
-        name = name[:-3]
+    name = name.removesuffix(".py")
     name = name.replace(os.path.sep, ".")
     module_file: Path = Path(HF_MODULES_CACHE) / module_path
     with _HF_REMOTE_CODE_LOCK:

src/transformers/modelcard.py

@@ -794,8 +794,7 @@ def parse_log_history(log_history):
     if idx > 0:
         eval_results = {}
         for key, value in log_history[idx].items():
-            if key.startswith("eval_"):
-                key = key[5:]
+            key = key.removeprefix("eval_")
             if key not in ["runtime", "samples_per_second", "steps_per_second", "epoch", "step"]:
                 camel_cased_key = " ".join([part.capitalize() for part in key.split("_")])
                 eval_results[camel_cased_key] = value

src/transformers/modeling_utils.py

@@ -5632,7 +5632,7 @@ def retrieve_modules_from_names(self, names, add_prefix=False, remove_prefix=Fal
         for name, module in self.named_modules():
             if remove_prefix:
                 _prefix = f"{self.base_model_prefix}."
-                name = name[len(_prefix) :] if name.startswith(_prefix) else name
+                name = name.removeprefix(_prefix)
             elif add_prefix:
                 name = ".".join([self.base_model_prefix, name]) if len(name) > 0 else self.base_model_prefix
 
@@ -5952,7 +5952,7 @@ def _adjust_missing_and_unexpected_keys(
         # in the warnings. For missing keys, we should show the prefix in the warning as it's part of the final model
         if loading_task_model_from_base_state_dict:
             _prefix = f"{self.base_model_prefix}."
-            unexpected_keys = [k[len(_prefix) :] if k.startswith(_prefix) else k for k in unexpected_keys]
+            unexpected_keys = [k.removeprefix(_prefix) for k in unexpected_keys]
 
         return missing_keys, unexpected_keys
 

src/transformers/models/auto/image_processing_auto.py

@@ -566,9 +566,7 @@ def from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs):
                     )
                 image_processor_class = get_image_processor_class_from_name(image_processor_type)
             else:
-                image_processor_type_slow = (
-                    image_processor_type[:-4] if image_processor_type.endswith("Fast") else image_processor_type
-                )
+                image_processor_type_slow = image_processor_type.removesuffix("Fast")
                 image_processor_class = get_image_processor_class_from_name(image_processor_type_slow)
                 if image_processor_class is None and image_processor_type.endswith("Fast"):
                     raise ValueError(

src/transformers/models/chinese_clip/convert_chinese_clip_original_pytorch_to_hf.py

@@ -105,7 +105,7 @@ def convert_chinese_clip_checkpoint(checkpoint_path, pytorch_dump_folder_path, c
     hf_model = ChineseCLIPModel(config).eval()
 
     pt_weights = torch.load(checkpoint_path, map_location="cpu", weights_only=True)["state_dict"]
-    pt_weights = {(name[7:] if name.startswith("module.") else name): value for name, value in pt_weights.items()}
+    pt_weights = {(name.removeprefix("module.")): value for name, value in pt_weights.items()}
 
     copy_text_model_and_projection(hf_model, pt_weights)
     copy_vision_model_and_projection(hf_model, pt_weights)

src/transformers/models/longt5/modeling_longt5.py

@@ -1272,8 +1272,7 @@ def dummy_inputs(self):
 
     def _try_load_missing_tied_module(self, key):
         module = self
-        if key.endswith(".weight"):
-            key = key[: -len(".weight")]
+        key = key.removesuffix(".weight")
         for sub_key in key.split("."):
             if not hasattr(module, sub_key):
                 return

src/transformers/models/rag/retrieval_rag.py

@@ -509,10 +509,7 @@ def postprocess_docs(self, docs, input_strings, prefix, n_docs, return_tensors=N
         def cat_input_and_doc(doc_title, doc_text, input_string, prefix):
             # TODO(Patrick): if we train more RAG models, I want to put the input first to take advantage of effortless truncation
             # TODO(piktus): better handling of truncation
-            if doc_title.startswith('"'):
-                doc_title = doc_title[1:]
-            if doc_title.endswith('"'):
-                doc_title = doc_title[:-1]
+            doc_title = doc_title.removeprefix('"').removesuffix('"')
             if prefix is None:
                 prefix = ""
             out = (prefix + doc_title + self.config.title_sep + doc_text + self.config.doc_sep + input_string).replace(

src/transformers/utils/auto_docstring.py

@@ -1215,8 +1215,7 @@ def get_checkpoint_from_config_class(config_class):
     # For example, `('google-bert/bert-base-uncased', 'https://huggingface.co/google-bert/bert-base-uncased')`
     for ckpt_name, ckpt_link in checkpoints:
         # allow the link to end with `/`
-        if ckpt_link.endswith("/"):
-            ckpt_link = ckpt_link[:-1]
+        ckpt_link = ckpt_link.removesuffix("/")
 
         # verify the checkpoint name corresponds to the checkpoint link
         ckpt_link_from_name = f"https://huggingface.co/{ckpt_name}"