[V0 Deprecation] Remove V0 sampling metadata (#25345)

Signed-off-by: Woosuk Kwon <woosuk@thinkingmachines.ai>

vllm/model_executor/models/ovis2_5.py

@@ -19,7 +19,6 @@ from vllm.model_executor.models.siglip2navit import Siglip2NavitModel
 from vllm.model_executor.models.utils import (AutoWeightsLoader, flatten_bn,
                                               init_vllm_registered_model,
                                               maybe_prefix)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems)
@@ -630,9 +629,8 @@ class Ovis2_5(nn.Module, SupportsMultiModal, SupportsPP):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.llm.compute_logits(hidden_states, sampling_metadata)
+        logits = self.llm.compute_logits(hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/paligemma.py

@@ -9,7 +9,6 @@ from transformers import BatchFeature, PaliGemmaConfig
 
 from vllm.config import VllmConfig
 from vllm.logger import init_logger
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalInputs, MultiModalKwargsItems,
@@ -403,10 +402,8 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsMultiModal,
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/persimmon.py

@@ -44,7 +44,6 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import SupportsPP
@@ -334,10 +333,8 @@ class PersimmonForCausalLM(nn.Module, SupportsPP):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/phi.py

@@ -59,7 +59,6 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import SupportsLoRA, SupportsPP
@@ -346,10 +345,9 @@ class PhiForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
         logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata, self.lm_head.bias)
+                                       self.lm_head.bias)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/phi3v.py

@@ -29,7 +29,6 @@ from vllm.logger import init_logger
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     VocabParallelEmbedding)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems)
@@ -681,10 +680,8 @@ class Phi3VForCausalLM(nn.Module, SupportsMultiModal, SupportsPP,
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/phi4_multimodal.py

@@ -27,7 +27,6 @@ from vllm.model_executor.layers.linear import (ColumnParallelLinear,
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.models.module_mapping import MultiModelKeys
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems, NestedTensors)
@@ -1451,10 +1450,8 @@ class Phi4MultimodalForCausalLM(nn.Module, SupportsLoRA, SupportsMultiModal):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/phi4flash.py

@@ -29,7 +29,6 @@ from vllm.model_executor.models.interfaces import (HasInnerState, IsHybrid,
                                                    SupportsV0Only)
 from vllm.model_executor.models.mamba_cache import (MambaCacheManager,
                                                     MambaCacheParams)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .utils import make_layers, maybe_prefix
@@ -695,12 +694,10 @@ class Phi4FlashForCausalLM(nn.Module, HasInnerState, IsHybrid, SupportsV0Only):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
         processed_logits = self.logits_processor(
             self.lm_head,
             hidden_states,
-            sampling_metadata,
             self.embedding_bias,
         )
         return processed_logits

vllm/model_executor/models/phi4mm.py

@@ -18,7 +18,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
     DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead)
 from vllm.model_executor.models.llama import LlamaModel
 from vllm.model_executor.models.module_mapping import MultiModelKeys
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems, NestedTensors)
@@ -1257,10 +1256,8 @@ class Phi4MMForCausalLM(nn.Module, SupportsLoRA, SupportsMultiModal):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/phimoe.py

@@ -47,7 +47,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
     DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead, VocabParallelEmbedding)
 from vllm.model_executor.model_loader.weight_utils import (
     default_weight_loader, maybe_remap_kv_scale_name)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import SupportsLoRA, SupportsPP
@@ -667,10 +666,8 @@ class PhiMoEForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
                                    inputs_embeds)
         return hidden_states
 
-    def compute_logits(self, hidden_states: torch.Tensor,
-                       sampling_metadata: SamplingMetadata) -> torch.Tensor:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+    def compute_logits(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/pixtral.py

@@ -32,7 +32,6 @@ from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalKwargsItems
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalUUIDDict, NestedTensors)
@@ -480,10 +479,8 @@ class PixtralForConditionalGeneration(nn.Module, SupportsMultiModal,
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
 

vllm/model_executor/models/plamo2.py

@@ -52,7 +52,6 @@ from vllm.model_executor.models.mamba_cache import (MambaCacheManager,
 from vllm.model_executor.models.utils import (
     is_pp_missing_parameter, make_empty_intermediate_tensors_factory,
     make_layers, maybe_prefix)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.model_executor.utils import set_weight_attrs
 from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
@@ -1022,10 +1021,8 @@ class Plamo2ForCausalLM(torch.nn.Module, HasInnerState, SupportsPP, IsHybrid):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):

vllm/model_executor/models/qwen.py

@@ -30,7 +30,6 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import SupportsLoRA, SupportsPP
@@ -282,10 +281,8 @@ class QWenBaseModel(nn.Module):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/qwen2.py

@@ -49,7 +49,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
 from vllm.model_executor.model_loader.weight_utils import (
     default_weight_loader, maybe_remap_kv_scale_name)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.config import is_interleaved
 
@@ -510,10 +509,8 @@ class Qwen2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsEagle3):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/qwen2_5_omni_thinker.py

@@ -50,7 +50,6 @@ from vllm.model_executor.models.qwen2_5_vl import (
 from vllm.model_executor.models.qwen2_audio import (
     Qwen2AudioProcessingInfo, _get_feat_extract_output_lengths)
 from vllm.model_executor.models.qwen2_vl import Qwen2VLMultiModalDataParser
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (ImageItem, ModalityData,
                                     MultiModalDataDict, MultiModalFieldConfig,
@@ -955,10 +954,8 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/qwen2_5_vl.py

@@ -43,7 +43,6 @@ from vllm.config import VllmConfig
 from vllm.distributed import parallel_state
 from vllm.distributed import utils as dist_utils
 from vllm.logger import init_logger
-from vllm.model_executor import SamplingMetadata
 from vllm.model_executor.layers.activation import get_act_and_mul_fn
 from vllm.model_executor.layers.layernorm import RMSNorm
 # yapf: disable
@@ -1256,10 +1255,8 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal,
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/qwen2_audio.py

@@ -34,7 +34,6 @@ from transformers.models.qwen2_audio import (Qwen2AudioConfig,
 from transformers.models.whisper import WhisperFeatureExtractor
 
 from vllm.config import VllmConfig
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (AudioItem, ModalityData,
                                     MultiModalDataDict, MultiModalFieldConfig,
@@ -481,10 +480,8 @@ class Qwen2AudioForConditionalGeneration(nn.Module, SupportsMultiModal,
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/qwen2_moe.py

@@ -51,7 +51,6 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import SupportsLoRA, SupportsPP
@@ -546,10 +545,8 @@ class Qwen2MoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/qwen2_vl.py

@@ -46,7 +46,6 @@ from vllm.config import VllmConfig
 from vllm.distributed import parallel_state, tensor_model_parallel_all_gather
 from vllm.distributed import utils as dist_utils
 from vllm.logger import init_logger
-from vllm.model_executor import SamplingMetadata
 from vllm.model_executor.layers.activation import QuickGELU
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                RowParallelLinear)
@@ -1527,10 +1526,8 @@ class Qwen2VLForConditionalGeneration(nn.Module, SupportsMultiModal,
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.language_model.compute_logits(hidden_states,
-                                                  sampling_metadata)
+        return self.language_model.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/qwen3.py

@@ -41,7 +41,6 @@ from vllm.model_executor.layers.logits_processor import LogitsProcessor
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import SupportsEagle3, SupportsLoRA, SupportsPP
@@ -328,10 +327,8 @@ class Qwen3ForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsEagle3):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/qwen3_moe.py

@@ -54,7 +54,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
 from vllm.model_executor.model_loader.weight_utils import (
     default_weight_loader, maybe_remap_kv_scale_name)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import MixtureOfExperts, SupportsLoRA, SupportsPP
@@ -690,10 +689,8 @@ class Qwen3MoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA,
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,