Optimizing the performance of think length limit using custom operators (#4279)

* delete impl

* delete min_length&max_length

* support limit thinking content strategy

* fix

* fix

* fix

* update

* fix set_value_by_flags_and_idx

* fix

* fix

* fix

* fix

* update

* fix

* fix

* fix typo

* fix ci

* fix

* fix

* support mtp

* fix

* fix

* update

* update

fastdeploy/model_executor/models/interfaces_base.py

@@ -26,33 +26,10 @@ T = TypeVar("T", default=paddle.Tensor)
 T_co = TypeVar("T_co", default=paddle.Tensor, covariant=True)
 
 
-def is_text_generation_model(model_cls: Type[nn.Layer]) -> bool:
-    from .model_base import ModelForCasualLM
-
-    return issubclass(model_cls, ModelForCasualLM)
-
-
 def is_pooling_model(model_cls: Type[nn.Layer]) -> bool:
     return getattr(model_cls, "is_pooling_model", False)
 
 
-def is_multimodal_model(class_name: str) -> bool:
-    multimodal_indicators = ["VL", "Vision", "ConditionalGeneration"]
-    return any(indicator in class_name for indicator in multimodal_indicators)
-
-
-def determine_model_category(class_name: str):
-    from fastdeploy.model_executor.models.model_base import ModelCategory
-
-    if any(pattern in class_name for pattern in ["VL", "Vision", "ConditionalGeneration"]):
-        return ModelCategory.MULTIMODAL
-    elif any(pattern in class_name for pattern in ["Embedding", "ForSequenceClassification"]):
-        return ModelCategory.EMBEDDING
-    elif any(pattern in class_name for pattern in ["Reward"]):
-        return ModelCategory.REWARD
-    return ModelCategory.TEXT_GENERATION
-
-
 def get_default_pooling_type(model_cls: Type[nn.Layer] = None) -> str:
     if model_cls is not None:
         return getattr(model_cls, "default_pooling_type", "LAST")