Remove deprecation cycle for candidate generators (#50034)

remove

Author: Cyrilvallez

src/transformers/generation/candidate_generator.py

@@ -22,7 +22,6 @@
 
 from ..pytorch_utils import prune_linear_layer
 from ..utils import ModelOutput, is_sklearn_available
-from ..utils.deprecation import deprecate_kwarg
 from .configuration_utils import GenerationConfig
 from .logits_process import LogitsProcessorList, MinLengthLogitsProcessor, SuppressTokensLogitsProcessor
 
@@ -1250,21 +1249,11 @@ class SinglePositionMultiTokenCandidateGenerator(CandidateGenerator):
         "return_shared_kv_states": True,
     }
 
-    @deprecate_kwarg("input_ids", version="5.20.0")
-    @deprecate_kwarg("model_kwargs", version="5.20.0")
-    @deprecate_kwarg("inputs_tensor", version="5.20.0")
-    @deprecate_kwarg("logits_processor", version="5.20.0")
-    @deprecate_kwarg("eos_token_id", version="5.20.0")
     def __init__(
         self,
         assistant_model: "PreTrainedModel",
         target_model_input_embeddings: nn.Embedding,
         generation_config: "GenerationConfig",
-        input_ids: torch.LongTensor | None = None,
-        model_kwargs: dict | None = None,
-        inputs_tensor: torch.Tensor | None = None,
-        logits_processor: Optional["LogitsProcessorList"] = None,
-        eos_token_id: int | list[int] | torch.Tensor | None = None,
         **kwargs,
     ):
         if (