remove unused `use_cache` in config classes (#20844)

remove unused use_cache in config classes
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

src/transformers/models/canine/configuration_canine.py

@@ -104,7 +104,6 @@ class CanineConfig(PretrainedConfig):
         type_vocab_size=16,
         initializer_range=0.02,
         layer_norm_eps=1e-12,
-        use_cache=True,
         pad_token_id=0,
         bos_token_id=0xE000,
         eos_token_id=0xE001,
@@ -128,7 +127,6 @@ class CanineConfig(PretrainedConfig):
         self.initializer_range = initializer_range
         self.type_vocab_size = type_vocab_size
         self.layer_norm_eps = layer_norm_eps
-        self.use_cache = use_cache
 
         # Character config:
         self.downsampling_rate = downsampling_rate

src/transformers/models/lilt/configuration_lilt.py

@@ -70,9 +70,6 @@ class LiltConfig(PretrainedConfig):
             [Self-Attention with Relative Position Representations (Shaw et al.)](https://arxiv.org/abs/1803.02155).
             For more information on `"relative_key_query"`, please refer to *Method 4* in [Improve Transformer Models
             with Better Relative Position Embeddings (Huang et al.)](https://arxiv.org/abs/2009.13658).
-        use_cache (`bool`, *optional*, defaults to `True`):
-            Whether or not the model should return the last key/values attentions (not used by all models). Only
-            relevant if `config.is_decoder=True`.
         classifier_dropout (`float`, *optional*):
             The dropout ratio for the classification head.
         channel_shrink_ratio (`int`, *optional*, defaults to 4):
@@ -111,7 +108,6 @@ class LiltConfig(PretrainedConfig):
         layer_norm_eps=1e-12,
         pad_token_id=0,
         position_embedding_type="absolute",
-        use_cache=True,
         classifier_dropout=None,
         channel_shrink_ratio=4,
         max_2d_position_embeddings=1024,
@@ -132,7 +128,6 @@ class LiltConfig(PretrainedConfig):
         self.initializer_range = initializer_range
         self.layer_norm_eps = layer_norm_eps
         self.position_embedding_type = position_embedding_type
-        self.use_cache = use_cache
         self.classifier_dropout = classifier_dropout
         self.channel_shrink_ratio = channel_shrink_ratio
         self.max_2d_position_embeddings = max_2d_position_embeddings

src/transformers/models/longformer/configuration_longformer.py

@@ -92,9 +92,6 @@ class LongformerConfig(PretrainedConfig):
             [Self-Attention with Relative Position Representations (Shaw et al.)](https://arxiv.org/abs/1803.02155).
             For more information on `"relative_key_query"`, please refer to *Method 4* in [Improve Transformer Models
             with Better Relative Position Embeddings (Huang et al.)](https://arxiv.org/abs/2009.13658).
-        use_cache (`bool`, *optional*, defaults to `True`):
-            Whether or not the model should return the last key/values attentions (not used by all models). Only
-            relevant if `config.is_decoder=True`.
         classifier_dropout (`float`, *optional*):
             The dropout ratio for the classification head.
         attention_window (`int` or `List[int]`, *optional*, defaults to 512):
@@ -137,7 +134,6 @@ class LongformerConfig(PretrainedConfig):
         initializer_range: float = 0.02,
         layer_norm_eps: float = 1e-12,
         position_embedding_type: str = "absolute",
-        use_cache: bool = True,
         classifier_dropout: float = None,
         onnx_export: bool = False,
         **kwargs
@@ -162,7 +158,6 @@ class LongformerConfig(PretrainedConfig):
         self.initializer_range = initializer_range
         self.layer_norm_eps = layer_norm_eps
         self.position_embedding_type = position_embedding_type
-        self.use_cache = use_cache
         self.classifier_dropout = classifier_dropout
         self.onnx_export = onnx_export