sync v0.15.1 (models)

vllm/model_executor/models/kanana_v.py

@@ -732,7 +732,7 @@ class KananaVForConditionalGeneration(nn.Module, SupportsMultiModal, 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,
@@ -755,4 +755,4 @@ class KananaVForConditionalGeneration(nn.Module, SupportsMultiModal, 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/keye.py

@@ -1438,7 +1438,7 @@ class BaseKeyeModule(nn.Module, SupportsMultiModal):
 
     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,
@@ -1692,4 +1692,4 @@ class KeyeForConditionalGeneration(
         llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1)
         mrope_position_delta = (llm_positions.max() + 1 - len(input_tokens)).item()
 
-        return llm_positions, mrope_position_delta
+        return llm_positions, mrope_position_delta
\ No newline at end of file

vllm/model_executor/models/kimi_linear.py

@@ -506,7 +506,7 @@ class KimiLinearForCausalLM(
 
     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,
@@ -666,4 +666,4 @@ def get_spec_layer_idx_from_weight_name(
         for i in range(config.num_nextn_predict_layers):
             if weight_name.startswith(f"model.layers.{layer_idx + i}."):
                 return layer_idx + i
-    return None
+    return None
\ No newline at end of file

vllm/model_executor/models/kimi_vl.py

@@ -389,7 +389,7 @@ class KimiVLForConditionalGeneration(nn.Module, SupportsMultiModal, 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,
@@ -412,4 +412,4 @@ class KimiVLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP):
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
         loader = AutoWeightsLoader(self)
-        return loader.load_weights(weights)
+        return loader.load_weights(weights)
\ No newline at end of file

vllm/model_executor/models/lfm2.py

@@ -342,7 +342,7 @@ class Lfm2Model(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,
@@ -503,7 +503,7 @@ class Lfm2ForCausalLM(
 
     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,
@@ -523,4 +523,4 @@ class Lfm2ForCausalLM(
             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/lfm2_moe.py

@@ -457,7 +457,7 @@ class Lfm2MoeModel(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,
@@ -730,7 +730,7 @@ class Lfm2MoeForCausalLM(
 
     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,
@@ -753,4 +753,4 @@ class Lfm2MoeForCausalLM(
         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/lfm2_vl.py

@@ -769,7 +769,7 @@ class Lfm2VLForConditionalGeneration(
 
     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,
@@ -804,4 +804,4 @@ class Lfm2VLForConditionalGeneration(
             language_model="language_model",
             connector="multi_modal_projector",
             tower_model="vision_tower",
-        )
+        )
\ No newline at end of file

vllm/model_executor/models/llama.py

@@ -651,7 +651,7 @@ class LlamaForCausalLM(
 
     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/llava.py

@@ -662,7 +662,7 @@ class LlavaForConditionalGeneration(
 
     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,
@@ -853,4 +853,4 @@ class MantisMultiModalProcessor(LlavaMultiModalProcessor):
     dummy_inputs=LlavaDummyInputsBuilder,
 )
 class MantisForConditionalGeneration(LlavaForConditionalGeneration):
-    pass
+    pass
\ No newline at end of file

vllm/model_executor/models/llava_next.py

@@ -509,7 +509,7 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
 
     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,
@@ -577,4 +577,4 @@ model_executor.models.llava_next.LlavaNextProcessingInfo.get_num_image_tokens].
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
         loader = AutoWeightsLoader(self)
-        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/llava_next_video.py

@@ -426,7 +426,7 @@ class LlavaNextVideoForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
 
     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,
@@ -459,4 +459,4 @@ class LlavaNextVideoForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
             # This model doesn't support images for now
             ignore_unexpected_prefixes=["image_newline"],
         )
-        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/llava_onevision.py

@@ -887,7 +887,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
 
     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,
@@ -916,4 +916,4 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
         loader = AutoWeightsLoader(self)
-        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/longcat_flash.py

@@ -520,7 +520,7 @@ class FlashModel(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,
@@ -605,7 +605,7 @@ class LongcatFlashForCausalLM(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,
@@ -764,4 +764,4 @@ class LongcatFlashForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
                     self_attn.kv_a_layernorm.weight.data *= (
                         self.config.hidden_size / self.config.kv_lora_rank
                     ) ** 0.5
-        return loaded_params
+        return loaded_params
\ No newline at end of file

vllm/model_executor/models/longcat_flash_mtp.py

@@ -150,7 +150,7 @@ class LongCatFlashMTP(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,
@@ -345,4 +345,4 @@ class LongCatFlashMTP(nn.Module):
     ) -> int | None:
         if "model.mtp" in weight_name:
             return config.num_hidden_layers * 2
-        return None
+        return None
\ No newline at end of file

vllm/model_executor/models/mamba.py

@@ -142,7 +142,7 @@ class MambaModel(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,
@@ -225,7 +225,7 @@ class MambaForCausalLM(
 
     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,
@@ -279,4 +279,4 @@ class MambaForCausalLM(
 
     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/mamba2.py

@@ -137,7 +137,7 @@ class Mamba2Model(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,
@@ -267,7 +267,7 @@ class Mamba2ForCausalLM(
 
     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,
@@ -291,4 +291,4 @@ class Mamba2ForCausalLM(
 
     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/medusa.py

@@ -2,8 +2,6 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import os
-from typing import Iterable, List, Optional, Set, Tuple, Any, Dict
-
 from collections.abc import Iterable
 
 import torch
@@ -20,8 +18,6 @@ from .utils import maybe_prefix
 
 from vllm import _custom_ops as ops
 
-TOPK=10 # topk for sparse tree (10 is a placeholder and it is sufficient)
-
 
 class ResidualBlock(nn.Module):
     def __init__(self, config: VllmConfig, hidden_size: int, num_layers: int) -> None:

vllm/model_executor/models/midashenglm.py

@@ -796,7 +796,7 @@ class MiDashengLMModel(nn.Module, SupportsMultiModal, 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,
@@ -820,4 +820,4 @@ class MiDashengLMModel(nn.Module, SupportsMultiModal, 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/mimo.py

@@ -61,7 +61,7 @@ logger = init_logger(__name__)
 class MiMoModel(Qwen2Model):
     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,
@@ -185,4 +185,4 @@ class MiMoForCausalLM(Qwen2ForCausalLM, nn.Module):
     ) -> torch.Tensor | None:
         hidden_states = self.model.norm(hidden_states)
         logits = self.logits_processor(self.lm_head, hidden_states)
-        return logits
+        return logits
\ No newline at end of file

vllm/model_executor/models/mimo_mtp.py

@@ -169,7 +169,7 @@ class MiMoMTP(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,
@@ -291,4 +291,4 @@ class MiMoMTP(nn.Module):
             name = name.replace(
                 f"model.layers.{spec_layer}.", f"model.layers.{spec_layer}.mtp_block."
             )
-        return name
+        return name
\ No newline at end of file