Merge remote-tracking branch 'origin/v0.15.1-dev' into v0.15.1-dev

# Conflicts:
#	vllm/model_executor/layers/fused_moe/config.py
#	vllm/model_executor/layers/fused_moe/layer.py
#	vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_marlin.py

vllm/model_executor/models/mixtral.py

@@ -347,7 +347,7 @@ class MixtralModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -608,7 +608,7 @@ class MixtralForCausalLM(nn.Module, SupportsLoRA, SupportsPP, MixtureOfExperts):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/mllama4.py

@@ -901,7 +901,7 @@ class Llama4ForConditionalGeneration(
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -1161,4 +1161,4 @@ class Llama4ForConditionalGeneration(
             language_model="language_model",
             connector="multi_modal_projector.",
             tower_model="vision_model.",
-        )
+        )
\ No newline at end of file

vllm/model_executor/models/mlp_speculator.py

@@ -3,7 +3,6 @@
 
 import os
 import math
-from typing import Iterable, List, Set, Tuple, Optional
 from collections.abc import Iterable
 
 import torch
@@ -17,7 +16,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm import _custom_ops as ops
-from vllm.distributed import tensor_model_parallel_all_gather, tensor_model_parallel_gather
 from vllm import envs
 
 from .utils import maybe_prefix

vllm/model_executor/models/modernbert.py

@@ -54,11 +54,12 @@ class ModernBertEmbeddings(nn.Module):
         input_ids: torch.Tensor,
         inputs_embeds: torch.Tensor | None = None,
     ) -> torch.Tensor:
-        if inputs_embeds is None:
+        if inputs_embeds is not None:
+            return self.norm(inputs_embeds)
+        else:
             inputs_embeds = self.tok_embeddings(input_ids)
-
-        embeddings = self.norm(inputs_embeds)
-        return embeddings
+            embeddings = self.norm(inputs_embeds)
+            return embeddings
 
 
 class ModernBertAttention(nn.Module):
@@ -454,4 +455,4 @@ class ModernBertForTokenClassification(nn.Module):
         )
         hidden_states = self.head(hidden_states)
         hidden_states = hidden_states.to(self.head_dtype)
-        return self.classifier(hidden_states)
+        return self.classifier(hidden_states)
\ No newline at end of file

vllm/model_executor/models/molmo.py

@@ -871,7 +871,7 @@ class MolmoModel(nn.Module, SupportsQuant):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -1591,4 +1591,4 @@ def _get_weights_with_merged_embedding(
         [embedding_weights["embedding"], embedding_weights["new_embedding"]],
         dim=0,
     )
-    yield ("model.embed_tokens.weight", embedding_weights)
+    yield ("model.embed_tokens.weight", embedding_weights)
\ No newline at end of file

vllm/model_executor/models/molmo2.py

@@ -1217,7 +1217,7 @@ class Molmo2TextModel(nn.Module, SupportsQuant):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -2805,4 +2805,4 @@ def _get_weights_with_merged_embedding(
         [embedding_weights["embedding"], embedding_weights["new_embedding"]],
         dim=0,
     )
-    yield ("model.embed_tokens.weight", embedding_weights)
+    yield ("model.embed_tokens.weight", embedding_weights)
\ No newline at end of file

vllm/model_executor/models/mpt.py

@@ -253,7 +253,7 @@ class MPTModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         position_ids: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -313,7 +313,7 @@ class MPTForCausalLM(nn.Module, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -332,4 +332,4 @@ class MPTForCausalLM(nn.Module, SupportsPP):
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
         loader = AutoWeightsLoader(self)
-        return loader.load_weights(weights)
+        return loader.load_weights(weights)
\ No newline at end of file

vllm/model_executor/models/nano_nemotron_vl.py

@@ -1917,7 +1917,7 @@ class NemotronH_Nano_VL_V2(
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,

vllm/model_executor/models/nemotron.py

@@ -477,7 +477,7 @@ class NemotronForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -496,4 +496,4 @@ class NemotronForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
         loader = AutoWeightsLoader(self)
-        return loader.load_weights(weights)
+        return loader.load_weights(weights)
\ No newline at end of file

vllm/model_executor/models/nemotron_h.py

@@ -601,7 +601,7 @@ class NemotronHModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -887,7 +887,7 @@ class NemotronHForCausalLM(
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -908,4 +908,4 @@ class NemotronHForCausalLM(
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
         loader = AutoWeightsLoader(self, skip_prefixes=["mtp"])
-        return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
+        return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
\ No newline at end of file

vllm/model_executor/models/nemotron_nas.py

@@ -449,7 +449,7 @@ class DeciLMForCausalLM(nn.Module, SupportsLoRA, SupportsPP, HasNoOps):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -471,4 +471,4 @@ class DeciLMForCausalLM(nn.Module, SupportsLoRA, SupportsPP, HasNoOps):
             self,
             skip_prefixes=(["lm_head."] if self.config.tie_word_embeddings else None),
         )
-        return loader.load_weights(weights)
+        return loader.load_weights(weights)
\ No newline at end of file

vllm/model_executor/models/nemotron_parse.py

@@ -289,7 +289,7 @@ class MBartDecoderNoPos(nn.Module):
 
     def forward(
         self,
-        decoder_input_ids: torch.Tensor | None,
+        decoder_input_ids: torch.Tensor,
         *,
         encoder_hidden_states: torch.Tensor | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -897,7 +897,7 @@ class NemotronParseForConditionalGeneration(nn.Module, SupportsMultiModal):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         encoder_outputs: list[torch.Tensor] | None = None,
         **kwargs,
@@ -957,4 +957,4 @@ class NemotronParseForConditionalGeneration(nn.Module, SupportsMultiModal):
         # Load encoder weights
         self.encoder.load_weights(encoder_weights)
         # Load decoder weights
-        self.decoder.load_weights(decoder_weights)
+        self.decoder.load_weights(decoder_weights)
\ No newline at end of file

vllm/model_executor/models/nemotron_vl.py

@@ -597,7 +597,7 @@ class LlamaNemotronVLChatModel(nn.Module, SupportsMultiModal, SupportsPP, Suppor
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -642,4 +642,4 @@ class LlamaNemotronVLChatModel(nn.Module, SupportsMultiModal, SupportsPP, Suppor
             language_model="language_model",
             connector="mlp1",
             tower_model="vision_model",
-        )
+        )
\ No newline at end of file

vllm/model_executor/models/olmo.py

@@ -271,7 +271,7 @@ class OlmoModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -382,7 +382,7 @@ class OlmoForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -409,4 +409,4 @@ class OlmoForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
                 ["lm_head.weight"] if self.config.tie_word_embeddings else None
             ),
         )
-        return loader.load_weights(weights)
+        return loader.load_weights(weights)
\ No newline at end of file

vllm/model_executor/models/olmo2.py

@@ -309,7 +309,7 @@ class Olmo2Model(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -424,7 +424,7 @@ class Olmo2ForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -451,4 +451,4 @@ class Olmo2ForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
                 ["lm_head.weight"] if self.config.tie_word_embeddings else None
             ),
         )
-        return loader.load_weights(weights)
+        return loader.load_weights(weights)
\ No newline at end of file

vllm/model_executor/models/olmoe.py

@@ -300,7 +300,7 @@ class OlmoeModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -476,7 +476,7 @@ class OlmoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -495,4 +495,4 @@ class OlmoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
         return loader.load_weights(weights)
 
     def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
-        return self.model.get_expert_mapping()
+        return self.model.get_expert_mapping()
\ No newline at end of file

vllm/model_executor/models/openpangu.py

@@ -1056,7 +1056,7 @@ class OpenPanguModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -1286,7 +1286,7 @@ class OpenPanguModelBase(nn.Module, SupportsPP, SupportsLoRA):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -1375,4 +1375,4 @@ class PanguUltraMoEForCausalLM(OpenPanguMoEModel):
 
 
 class PanguProMoEV2ForCausalLM(OpenPanguMoEModel):
-    pass
+    pass
\ No newline at end of file

vllm/model_executor/models/openpangu_mtp.py

@@ -104,7 +104,7 @@ class OpenPanguMTP(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         hidden_states: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
@@ -262,4 +262,4 @@ class OpenPanguMTP(nn.Module):
         elif shared_weight:
             # treat shared weights as top level weights
             name = name.replace(f"model.layers.{spec_layer}.", "model.")
-        return name
+        return name
\ No newline at end of file

vllm/model_executor/models/opt.py

@@ -267,7 +267,7 @@ class OPTDecoder(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -316,7 +316,7 @@ class OPTModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -399,7 +399,7 @@ class OPTForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -423,4 +423,4 @@ class OPTForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
                 ["lm_head.weight"] if self.config.tie_word_embeddings else None
             ),
         )
-        return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
+        return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
\ No newline at end of file

vllm/model_executor/models/orion.py

@@ -253,7 +253,7 @@ class OrionModel(nn.Module):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None,
         inputs_embeds: torch.Tensor | None = None,
@@ -343,7 +343,7 @@ class OrionForCausalLM(nn.Module, SupportsPP):
 
     def forward(
         self,
-        input_ids: torch.Tensor | None,
+        input_ids: torch.Tensor,
         positions: torch.Tensor,
         intermediate_tensors: IntermediateTensors | None = None,
         inputs_embeds: torch.Tensor | None = None,
@@ -362,4 +362,4 @@ class OrionForCausalLM(nn.Module, SupportsPP):
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
         loader = AutoWeightsLoader(self)
-        return loader.load_weights(weights)
+        return loader.load_weights(weights)
\ No newline at end of file