transformers 4.57.2__py3-none-any.whl → 4.57.4__py3-none-any.whl

transformers/__init__.py

@@ -18,7 +18,7 @@
 # to defer the actual importing for when the objects are requested. This way `import transformers` provides the names
 # in the namespace without actually importing anything (and especially none of the backends).
 
-__version__ = "4.57.2"
+__version__ = "4.57.4"
 
 from pathlib import Path
 from typing import TYPE_CHECKING

transformers/generation/utils.py

@@ -2379,9 +2379,11 @@ class GenerationMixin(ContinuousMixin):
             generation_config, use_model_defaults, **kwargs
         )
         generation_mode = generation_config.get_generation_mode(assistant_model)
+        deprecated_mode_repo = self._get_deprecated_gen_repo(generation_mode, trust_remote_code, custom_generate)
+
         if isinstance(custom_generate, Callable):
             decoding_method = custom_generate
-        else:
+        elif deprecated_mode_repo is None:
             # type() required to access the unbound class-level method
             decoding_method = getattr(type(self), GENERATION_MODES_MAPPING[generation_mode])
 
@@ -2392,7 +2394,7 @@ class GenerationMixin(ContinuousMixin):
         # NOTE: This must come after initializing generation_config, since we need it to determine if this is a deprecated mode.
         # It must also be before any preparation steps, since Hub repos expect to be loaded before preparation steps.
         # TODO joao, manuel: remove this in v4.62.0
-        if deprecated_mode_repo := self._get_deprecated_gen_repo(generation_mode, trust_remote_code, custom_generate):
+        if deprecated_mode_repo is not None:
             return GenerationMixin.generate(
                 self,
                 inputs=inputs,

transformers/models/apertus/modeling_apertus.py

@@ -339,7 +339,7 @@ class ApertusModel(ApertusPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/arcee/modeling_arcee.py

@@ -344,7 +344,7 @@ class ArceeModel(ArceePreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/aria/modeling_aria.py

@@ -721,7 +721,7 @@ class AriaTextModel(AriaTextPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/audio_spectrogram_transformer/modeling_audio_spectrogram_transformer.py

@@ -366,7 +366,7 @@ class ASTModel(ASTPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/aya_vision/modeling_aya_vision.py

@@ -263,7 +263,7 @@ class AyaVisionModel(AyaVisionPreTrainedModel):
             )
         return special_image_mask
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/aya_vision/modular_aya_vision.py

@@ -162,7 +162,7 @@ class AyaVisionModel(LlavaModel):
         image_features = self.multi_modal_projector(selected_image_feature)
         return image_features
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/bitnet/modeling_bitnet.py

@@ -343,7 +343,7 @@ class BitNetModel(BitNetPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/blip_2/modeling_blip_2.py

@@ -1007,7 +1007,7 @@ class Blip2QFormerModel(Blip2PreTrainedModel):
         extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
         return extended_attention_mask
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/blt/modeling_blt.py

@@ -577,7 +577,7 @@ class BltLocalDecoder(BltPreTrainedModel):
 
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -1047,7 +1047,7 @@ class BltModel(BltPreTrainedModel):
             self.patcher = None
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/blt/modular_blt.py

@@ -536,7 +536,7 @@ class BltLocalDecoder(BltPreTrainedModel):
 
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -799,7 +799,7 @@ class BltModel(BltPreTrainedModel):
             self.patcher = None
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/cohere/modeling_cohere.py

@@ -376,7 +376,7 @@ class CohereModel(CoherePreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/cohere2/modeling_cohere2.py

@@ -351,7 +351,7 @@ class Cohere2Model(Cohere2PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/cohere2_vision/modeling_cohere2_vision.py

@@ -213,7 +213,7 @@ class Cohere2VisionModel(Cohere2VisionPreTrainedModel):
             )
         return special_image_mask
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,
@@ -306,7 +306,7 @@ class Cohere2VisionForConditionalGeneration(Cohere2VisionPreTrainedModel, Genera
     def multi_modal_projector(self):
         return self.model.multi_modal_projector
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/cohere2_vision/modular_cohere2_vision.py

@@ -107,7 +107,7 @@ class Cohere2VisionModel(AyaVisionModel):
         image_features = self.multi_modal_projector(selected_image_feature)
         return image_features
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,
@@ -160,7 +160,7 @@ class Cohere2VisionForConditionalGeneration(AyaVisionForConditionalGeneration):
     def get_image_features(self, pixel_values: torch.FloatTensor):
         return self.model.get_image_features(pixel_values=pixel_values)
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/csm/modeling_csm.py

@@ -409,7 +409,7 @@ class CsmDepthDecoderModel(CsmPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,
@@ -662,7 +662,7 @@ class CsmBackboneModel(CsmPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/csm/modular_csm.py

@@ -156,7 +156,7 @@ class CsmDepthDecoderModel(LlamaModel, CsmPreTrainedModel):
         self.embed_tokens = nn.Embedding((config.num_codebooks * config.vocab_size), config.backbone_hidden_size)
         self.inputs_embeds_projector = nn.Linear(config.backbone_hidden_size, config.hidden_size, bias=False)
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,
@@ -395,7 +395,7 @@ class CsmBackboneModel(LlamaModel):
         super().__init__(config)
         self.embed_tokens = CsmBackboneModelEmbeddings(config)
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(self, **super_kwargs):
         r"""

transformers/models/deepseek_v2/modeling_deepseek_v2.py

@@ -491,7 +491,7 @@ class DeepseekV2Model(DeepseekV2PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/deepseek_v3/modeling_deepseek_v3.py

@@ -539,7 +539,7 @@ class DeepseekV3Model(DeepseekV3PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/diffllama/modeling_diffllama.py

@@ -608,7 +608,7 @@ class DiffLlamaModel(DiffLlamaPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/dinov2/modeling_dinov2.py

@@ -624,7 +624,7 @@ class Dinov2Backbone(Dinov2PreTrainedModel, BackboneMixin):
     def get_input_embeddings(self) -> Dinov2PatchEmbeddings:
         return self.embeddings.patch_embeddings
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self, pixel_values: torch.Tensor, output_hidden_states: Optional[bool] = None, **kwargs

transformers/models/dinov2_with_registers/modeling_dinov2_with_registers.py

@@ -644,7 +644,7 @@ class Dinov2WithRegistersBackbone(Dinov2WithRegistersPreTrainedModel, BackboneMi
     def get_input_embeddings(self) -> Dinov2WithRegistersPatchEmbeddings:
         return self.embeddings.patch_embeddings
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/doge/modeling_doge.py

@@ -530,7 +530,7 @@ class DogeModel(DogePreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/dots1/modeling_dots1.py

@@ -454,7 +454,7 @@ class Dots1Model(Dots1PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/edgetam/modeling_edgetam.py

@@ -444,7 +444,7 @@ class EdgeTamVisionModel(EdgeTamPreTrainedModel):
 
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         pixel_values: Optional[torch.FloatTensor] = None,
@@ -1028,7 +1028,7 @@ class EdgeTamModel(EdgeTamPreTrainedModel):
         )
         return prompt_output
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/edgetam/modular_edgetam.py

@@ -208,7 +208,7 @@ class EdgeTamVisionModel(Sam2VisionModel):
     def get_input_embeddings(self):
         raise NotImplementedError("Can't get input embeddings from timm wrapper model")
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         pixel_values: Optional[torch.FloatTensor] = None,

transformers/models/efficientloftr/modeling_efficientloftr.py

@@ -680,7 +680,7 @@ class EfficientLoFTRModel(EfficientLoFTRPreTrainedModel):
 
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/emu3/modeling_emu3.py

@@ -1166,7 +1166,7 @@ class Emu3TextModel(Emu3PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/eomt/modeling_eomt.py

@@ -1087,7 +1087,7 @@ class EomtForUniversalSegmentation(EomtPreTrainedModel):
     def get_loss(self, loss_dict: dict[str, Tensor]) -> Tensor:
         return sum(loss_dict.values())
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/eomt/modular_eomt.py

@@ -492,7 +492,7 @@ class EomtForUniversalSegmentation(Mask2FormerForUniversalSegmentation):
 
         return attn_mask
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/ernie4_5/modeling_ernie4_5.py

@@ -342,7 +342,7 @@ class Ernie4_5Model(Ernie4_5PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/ernie4_5_moe/modeling_ernie4_5_moe.py

@@ -510,7 +510,7 @@ class Ernie4_5_MoeModel(Ernie4_5_MoePreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/ernie4_5_moe/modular_ernie4_5_moe.py

@@ -248,7 +248,7 @@ class Ernie4_5_MoeModel(Ernie4_5_MoePreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/esm/modeling_esm.py

@@ -678,7 +678,7 @@ class EsmModel(EsmPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/evolla/modeling_evolla.py

@@ -610,7 +610,7 @@ class EvollaSaProtProteinEncoder(EvollaSaProtPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.Tensor],
@@ -1397,7 +1397,7 @@ class EvollaModel(EvollaPreTrainedModel):
         self.embed_tokens = value
 
     @auto_docstring
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/evolla/modular_evolla.py

@@ -241,7 +241,7 @@ class EvollaSaProtProteinEncoder(EvollaSaProtPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.Tensor],
@@ -835,7 +835,7 @@ class EvollaModel(EvollaPreTrainedModel):
         self.embed_tokens = value
 
     @auto_docstring
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/exaone4/modeling_exaone4.py

@@ -352,7 +352,7 @@ class Exaone4Model(Exaone4PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/exaone4/modular_exaone4.py

@@ -364,7 +364,7 @@ class Exaone4Model(Exaone4PreTrainedModel, LlamaModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/flex_olmo/modeling_flex_olmo.py

@@ -417,7 +417,7 @@ class FlexOlmoModel(FlexOlmoPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/flex_olmo/modular_flex_olmo.py

@@ -277,7 +277,7 @@ class FlexOlmoPreTrainedModel(MixtralPreTrainedModel):
 # FlexOlmo model is identical to Mixtral model except:
 # - FlexOlmo does not use sliding window attention.
 class FlexOlmoModel(MixtralModel):
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/gemma/modeling_gemma.py

@@ -348,7 +348,7 @@ class GemmaModel(GemmaPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/gemma2/modeling_gemma2.py

@@ -379,7 +379,7 @@ class Gemma2Model(Gemma2PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/gemma3/modeling_gemma3.py

@@ -482,7 +482,7 @@ class Gemma3TextModel(Gemma3PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/glm/modeling_glm.py

@@ -358,7 +358,7 @@ class GlmModel(GlmPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/glm4/modeling_glm4.py

@@ -362,7 +362,7 @@ class Glm4Model(Glm4PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/glm4_moe/modeling_glm4_moe.py

@@ -476,7 +476,7 @@ class Glm4MoeModel(Glm4MoePreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/glm4v/modeling_glm4v.py

@@ -806,7 +806,7 @@ class Glm4vTextModel(Glm4vPreTrainedModel):
         self.post_init()
 
     @auto_docstring
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/glm4v/modular_glm4v.py

@@ -872,7 +872,7 @@ class Glm4vTextModel(Qwen2_5_VLTextModel):
         del self.has_sliding_layers
 
     @auto_docstring
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/glm4v_moe/modeling_glm4v_moe.py

@@ -922,7 +922,7 @@ class Glm4vMoeTextModel(Glm4vMoePreTrainedModel):
         self.post_init()
 
     @auto_docstring
-    @check_model_inputs()
+    @check_model_inputs
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

transformers/models/gpt_neox/modeling_gpt_neox.py

@@ -393,7 +393,7 @@ class GPTNeoXModel(GPTNeoXPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/gpt_oss/modeling_gpt_oss.py

@@ -456,7 +456,7 @@ class GptOssModel(GptOssPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/gpt_oss/modular_gpt_oss.py

@@ -386,7 +386,7 @@ class GptOssPreTrainedModel(LlamaPreTrainedModel):
 class GptOssModel(MixtralModel):
     _no_split_modules = ["GptOssDecoderLayer"]
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/granite/modeling_granite.py

@@ -375,7 +375,7 @@ class GraniteModel(GranitePreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,

transformers/models/helium/modeling_helium.py

@@ -343,7 +343,7 @@ class HeliumModel(HeliumPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @check_model_inputs()
+    @check_model_inputs
     @auto_docstring
     def forward(
         self,