Rename clashing method names for vLLM model protocol (#27583)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

vllm/model_executor/models/deepseek_eagle.py

@@ -73,7 +73,7 @@ class DeepseekV2Model(nn.Module):
         self.hnorm = RMSNorm(self.config.hidden_size, eps=self.config.rms_norm_eps)
         self.norm = RMSNorm(self.config.hidden_size, eps=self.config.rms_norm_eps)
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -222,8 +222,8 @@ class EagleDeepseekV3ForCausalLM(DeepseekV3ForCausalLM):
         self.num_moe_layers = self.config.num_hidden_layers
         self.set_moe_parameters()
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/deepseek_mtp.py

@@ -142,7 +142,7 @@ class DeepSeekMultiTokenPredictor(nn.Module):
         )
         self.logits_processor = LogitsProcessor(config.vocab_size)
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -206,8 +206,8 @@ class DeepSeekMTP(nn.Module, SupportsPP, DeepseekV2MixtureOfExperts):
                 self.moe_layers.append(layer.mlp.experts)
         self.extract_moe_parameters(example_moe)
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/deepseek_ocr.py

@@ -557,9 +557,7 @@ class DeepseekOCRForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
     def get_language_model(self) -> torch.nn.Module:
         return self.language_model
 
-    def get_multimodal_embeddings(
-        self, **kwargs: object
-    ) -> MultiModalEmbeddings | None:
+    def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings | None:
         image_input = self._parse_and_validate_image_input(**kwargs)
         if image_input is None:
             return None

vllm/model_executor/models/deepseek_v2.py

@@ -1236,7 +1236,7 @@ class DeepseekV2Model(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -1250,7 +1250,7 @@ class DeepseekV2Model(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -1389,8 +1389,8 @@ class DeepseekV2ForCausalLM(
 
         self.extract_moe_parameters(example_moe)
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/deepseek_vl2.py

@@ -619,7 +619,7 @@ class DeepseekVLV2ForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
     def get_language_model(self) -> torch.nn.Module:
         return self.language_model
 
-    def get_multimodal_embeddings(self, **kwargs: object) -> MultiModalEmbeddings:
+    def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings:
         image_input = self._parse_and_validate_image_input(**kwargs)
         if image_input is None:
             return []

vllm/model_executor/models/dots1.py

@@ -398,7 +398,7 @@ class Dots1Model(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -412,7 +412,7 @@ class Dots1Model(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -541,8 +541,8 @@ class Dots1ForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
             self.model.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/dots_ocr.py

@@ -840,7 +840,7 @@ class DotsOCRForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA
     def get_language_model(self) -> torch.nn.Module:
         return self.language_model
 
-    def get_multimodal_embeddings(self, **kwargs: object) -> MultiModalEmbeddings:
+    def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings:
         image_input = self._parse_and_validate_image_input(**kwargs)
         if image_input is None:
             return []
@@ -858,8 +858,8 @@ class DotsOCRForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA
         if intermediate_tensors is not None:
             inputs_embeds = None
         elif inputs_embeds is None:
-            vision_embeddings = self.get_multimodal_embeddings(**kwargs)
-            inputs_embeds = self.get_input_embeddings(
+            vision_embeddings = self.embed_multimodal(**kwargs)
+            inputs_embeds = self.embed_input_ids(
                 input_ids,
                 vision_embeddings,
                 is_multimodal=input_ids == self.config.image_token_id,

vllm/model_executor/models/ernie45_moe.py

@@ -465,7 +465,7 @@ class Ernie4_5_MoeModel(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -479,7 +479,7 @@ class Ernie4_5_MoeModel(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -726,8 +726,8 @@ class Ernie4_5_MoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA, MixtureOfExpe
                 moe.n_redundant_experts = self.num_redundant_experts
                 moe.experts.update_expert_map()
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/ernie45_vl.py

@@ -1656,9 +1656,7 @@ class Ernie4_5_VLMoeForConditionalGeneration(
 
         return modalities
 
-    def get_multimodal_embeddings(
-        self, **kwargs: object
-    ) -> MultiModalEmbeddings | None:
+    def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings | None:
         modalities = self._parse_and_validate_multimodal_inputs(**kwargs)
         if not modalities:
             return None
@@ -1681,7 +1679,7 @@ class Ernie4_5_VLMoeForConditionalGeneration(
 
         return multimodal_embeddings
 
-    def get_input_embeddings(
+    def embed_input_ids(
         self,
         input_ids: torch.Tensor,
         multimodal_embeddings: MultiModalEmbeddings | None = None,
@@ -1694,9 +1692,9 @@ class Ernie4_5_VLMoeForConditionalGeneration(
 
         # This is to satisfy the type checker for each overload
         if multimodal_embeddings is None or is_multimodal is None:
-            return super().get_input_embeddings(input_ids)
+            return super().embed_input_ids(input_ids)
 
-        return super().get_input_embeddings(
+        return super().embed_input_ids(
             input_ids,
             multimodal_embeddings=multimodal_embeddings,
             is_multimodal=is_multimodal,

vllm/model_executor/models/ernie45_vl_moe.py

@@ -561,7 +561,7 @@ class Ernie4_5_VLMoeModel(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -577,7 +577,7 @@ class Ernie4_5_VLMoeModel(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -642,8 +642,8 @@ class Ernie4_5_VLMoeForCausalLM(nn.Module, SupportsPP):
             self.model.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/ernie_mtp.py

@@ -112,7 +112,7 @@ class ErnieMultiTokenPredictor(nn.Module):
         )
         self.logits_processor = LogitsProcessor(config.vocab_size)
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -160,8 +160,8 @@ class ErnieMTP(nn.Module, SupportsPP):
         if self.config.tie_word_embeddings:
             self.lm_head.weight = self.model.embed_tokens.weight
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/exaone.py

@@ -357,7 +357,7 @@ class ExaoneModel(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.wte(input_ids)
 
     def forward(
@@ -371,7 +371,7 @@ class ExaoneModel(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -512,8 +512,8 @@ class ExaoneForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             self.transformer.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/exaone4.py

@@ -344,7 +344,7 @@ class Exaone4Model(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -358,7 +358,7 @@ class Exaone4Model(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -498,8 +498,8 @@ class Exaone4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             self.model.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/falcon.py

@@ -399,7 +399,7 @@ class FalconModel(nn.Module):
             ["hidden_states"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.word_embeddings(input_ids)
 
     def forward(
@@ -413,7 +413,7 @@ class FalconModel(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
         else:
             hidden_states = intermediate_tensors["hidden_states"]
         for layer in islice(self.h, self.start_layer, self.end_layer):
@@ -515,8 +515,8 @@ class FalconForCausalLM(nn.Module, SupportsPP):
             self.transformer.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.transformer.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.transformer.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/falcon_h1.py

@@ -461,7 +461,7 @@ class FalconH1Model(nn.Module):
         else:
             self.final_layernorm = PPMissingLayer()
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -476,7 +476,7 @@ class FalconH1Model(nn.Module):
                 hidden_states = inputs_embeds * self.embedding_multiplier
             else:
                 hidden_states = (
-                    self.get_input_embeddings(input_ids) * self.embedding_multiplier
+                    self.embed_input_ids(input_ids) * self.embedding_multiplier
                 )
         else:
             assert intermediate_tensors is not None
@@ -601,8 +601,8 @@ class FalconH1ForCausalLM(
             self.model.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/fuyu.py

@@ -333,7 +333,7 @@ class FuyuForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
     def get_language_model(self) -> torch.nn.Module:
         return self.language_model
 
-    def get_multimodal_embeddings(self, **kwargs: object) -> MultiModalEmbeddings:
+    def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings:
         image_input = self._parse_and_validate_image_input(**kwargs)
         if image_input is None:
             return []

vllm/model_executor/models/gemma.py

@@ -293,7 +293,7 @@ class GemmaModel(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -307,7 +307,7 @@ class GemmaModel(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             hidden_states *= self.normalizer
             residual = None
         else:
@@ -396,8 +396,8 @@ class GemmaForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             self.model.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/gemma2.py

@@ -290,7 +290,7 @@ class Gemma2Model(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.embed_tokens(input_ids)
 
     def forward(
@@ -304,7 +304,7 @@ class Gemma2Model(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             hidden_states *= self.normalizer
             residual = None
         else:
@@ -409,8 +409,8 @@ class Gemma2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             self.model.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/gemma3.py

@@ -393,7 +393,7 @@ class Gemma3Model(nn.Module):
             ["hidden_states", "residual"], config.hidden_size
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         # NOTE(woosuk): Only apply the normalizer to the output of
         # vocab embedding. Don't apply it to the vision embedding.
         return self.embed_tokens(input_ids) * self.normalizer
@@ -410,7 +410,7 @@ class Gemma3Model(nn.Module):
             if inputs_embeds is not None:
                 hidden_states = inputs_embeds
             else:
-                hidden_states = self.get_input_embeddings(input_ids)
+                hidden_states = self.embed_input_ids(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -540,8 +540,8 @@ class Gemma3ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
             self.model.make_empty_intermediate_tensors
         )
 
-    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
-        return self.model.get_input_embeddings(input_ids)
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
 
     def forward(
         self,

vllm/model_executor/models/gemma3_mm.py

@@ -596,7 +596,7 @@ class Gemma3ForConditionalGeneration(
     def get_language_model(self) -> torch.nn.Module:
         return self.language_model
 
-    def get_multimodal_embeddings(self, **kwargs: object) -> MultiModalEmbeddings:
+    def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings:
         image_input = self._parse_and_validate_image_input(**kwargs)
         if image_input is None:
             return []