Merge tag 'v0.15.0rc1' into v0.15.0rc1-dev

vllm/model_executor/models/deepseek_ocr.py

@@ -562,7 +562,7 @@ class DeepseekOCRForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, Supports
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/deepseek_v2.py

@@ -1101,7 +1101,7 @@ class DeepseekV2Model(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -1284,7 +1284,7 @@ class DeepseekV2ForCausalLM(
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/deepseek_vl2.py

@@ -614,7 +614,7 @@ class DeepseekVLV2ForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/dots1.py

@@ -394,7 +394,7 @@ class Dots1Model(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -538,7 +538,7 @@ class Dots1ForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/dots_ocr.py

@@ -754,7 +754,7 @@ class DotsOCRForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/eagle2_5_vl.py

@@ -432,7 +432,7 @@ class Eagle2_5_VLForConditionalGeneration(
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -440,7 +440,6 @@ class Eagle2_5_VLForConditionalGeneration(
     ) -> IntermediateTensors:
         """Forward pass through the model."""
         if intermediate_tensors is not None:
-            input_ids = None
             inputs_embeds = None
 
         forward_kwargs = {

vllm/model_executor/models/ernie45_moe.py

@@ -466,7 +466,7 @@ class Ernie4_5_MoeModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -727,7 +727,7 @@ class Ernie4_5_MoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA, MixtureOfExpe
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/ernie45_vl.py

@@ -1650,7 +1650,7 @@ class Ernie4_5_VLMoeForConditionalGeneration(
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/ernie45_vl_moe.py

@@ -565,7 +565,7 @@ class Ernie4_5_VLMoeModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -646,7 +646,7 @@ class Ernie4_5_VLMoeForCausalLM(nn.Module, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/ernie_mtp.py

@@ -164,7 +164,7 @@ class ErnieMTP(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         hidden_states: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,

vllm/model_executor/models/exaone.py

@@ -496,7 +496,7 @@ class ExaoneForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/exaone4.py

@@ -490,7 +490,7 @@ class Exaone4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/exaone_moe.py

@@ -549,7 +549,7 @@ class ExaoneMoeForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/falcon.py

@@ -423,7 +423,7 @@ class FalconModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/falcon_h1.py

@@ -459,7 +459,7 @@ class FalconH1Model(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -602,7 +602,7 @@ class FalconH1ForCausalLM(
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/fuyu.py

@@ -340,7 +340,7 @@ class FuyuForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/gemma.py

@@ -297,7 +297,7 @@ class GemmaModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -400,7 +400,7 @@ class GemmaForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/gemma2.py

@@ -410,7 +410,7 @@ class Gemma2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/gemma3.py

@@ -494,7 +494,7 @@ class Gemma3ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/gemma3_mm.py

@@ -618,7 +618,7 @@ class Gemma3ForConditionalGeneration(
 
     def forward(
         self,
-        input_ids: torch.Tensor,
+        input_ids: torch.Tensor | None,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -656,3 +656,41 @@ class Gemma3ForConditionalGeneration(
             connector="multi_modal_projector",
             tower_model="vision_tower",
         )
+
+    def get_num_mm_encoder_tokens(self, num_image_tokens: int) -> int:
+        """
+        Calculate the number of tokens output by the vision encoder.
+
+        The vision encoder processes images into patch embeddings. For Gemma3,
+        the relationship between prompt placeholder tokens and actual vision
+        encoder output tokens depends on the patch grid size.
+
+        Args:
+            num_image_tokens: Number of image placeholder tokens in the prompt
+                              (typically mm_tokens_per_image per image)
+
+        Returns:
+            Number of tokens output by the vision encoder
+        """
+        # For Gemma3, the vision encoder outputs tokens_per_side x tokens_per_side
+        # tokens per image. Since num_image_tokens represents the number of
+        # connector output tokens (mm_tokens_per_image = 256), and tokens_per_side
+        # is sqrt(256) = 16, we need to account for the token expansion.
+        # Based on empirical testing, the multiplier of 16 works correctly.
+        return num_image_tokens * 16
+
+    def get_num_mm_connector_tokens(self, num_vision_tokens: int) -> int:
+        """
+        Calculate the number of tokens output by the multimodal connector.
+
+        The connector applies projection and normalization but maintains the
+        token count for Gemma3.
+
+        Args:
+            num_vision_tokens: Number of tokens from vision encoder
+
+        Returns:
+            Number of tokens after connector processing
+        """
+        # The Gemma3 connector maintains a 1:1 token mapping
+        return num_vision_tokens