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

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

vllm/model_executor/models/minicpm_eagle.py

@@ -39,7 +39,6 @@ from vllm.model_executor.layers.quantization import QuantizationConfig
 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
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import SupportsLoRA, SupportsPP
@@ -376,10 +375,8 @@ class EagleMiniCPMForCausalLM(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)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/minicpmv.py

@@ -50,7 +50,6 @@ from vllm.model_executor.models.minicpm import MiniCPMForCausalLM
 from vllm.model_executor.models.module_mapping import MultiModelKeys
 from vllm.model_executor.models.qwen2 import Qwen2ForCausalLM
 from vllm.model_executor.models.qwen3 import Qwen3ForCausalLM
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems, NestedTensors)
@@ -1194,9 +1193,8 @@ class MiniCPMVBaseModel(nn.Module, SupportsMultiModal, SupportsPP):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.llm.compute_logits(hidden_states, sampling_metadata)
+        return self.llm.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/minimax_text_01.py

@@ -41,7 +41,6 @@ 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
 from vllm.model_executor.models.utils import maybe_prefix
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import HasInnerState, IsHybrid
@@ -742,10 +741,8 @@ class MiniMaxText01ForCausalLM(nn.Module, HasInnerState, IsHybrid):
 
         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.float(),
-                                       sampling_metadata)
+    def compute_logits(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        logits = self.logits_processor(self.lm_head, hidden_states.float())
 
         return logits
 

vllm/model_executor/models/minimax_vl_01.py

@@ -14,7 +14,6 @@ from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.quantization import QuantizationConfig
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import MultiModalFieldConfig
 from vllm.sequence import IntermediateTensors
@@ -420,10 +419,8 @@ class MiniMaxVL01ForConditionalGeneration(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/mistral3.py

@@ -20,7 +20,6 @@ from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.quantization import QuantizationConfig
 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.cache import BaseMultiModalProcessorCache
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
@@ -606,10 +605,8 @@ class Mistral3ForConditionalGeneration(nn.Module, SupportsLoRA,
     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/mixtral.py

@@ -49,7 +49,6 @@ 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 MixtureOfExperts, SupportsLoRA, SupportsPP
@@ -594,10 +593,8 @@ class MixtralForCausalLM(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)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/mllama4.py

@@ -41,7 +41,6 @@ from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.model_loader.utils import initialize_model
 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
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems, NestedTensors)
@@ -856,10 +855,8 @@ class Llama4ForConditionalGeneration(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 separate_weights(
         self,

vllm/model_executor/models/molmo.py

@@ -26,7 +26,6 @@ from vllm.distributed import (get_pp_group, get_tensor_model_parallel_rank,
                               get_tensor_model_parallel_world_size,
                               split_tensor_along_last_dim,
                               tensor_model_parallel_all_gather)
-from vllm.model_executor import SamplingMetadata
 from vllm.model_executor.layers.activation import (MulAndSilu, QuickGELU,
                                                    SiluAndMul)
 from vllm.model_executor.layers.layernorm import RMSNorm
@@ -1527,10 +1526,8 @@ class MolmoForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA,
 
         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, torch.Tensor]]):

vllm/model_executor/models/mpt.py

@@ -25,7 +25,6 @@ from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     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
@@ -320,10 +319,8 @@ class MPTForCausalLM(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/nano_nemotron_vl.py

@@ -37,7 +37,6 @@ from vllm.model_executor.models.utils import (flatten_bn,
                                               init_vllm_registered_model,
                                               maybe_prefix,
                                               merge_multimodal_embeddings)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargs, MultiModalKwargsItems,
@@ -1192,10 +1191,8 @@ class NemotronH_Nano_VL(nn.Module, HasInnerState, IsHybrid,
     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]]):
         adapter_dict = dict(self.mlp1.named_parameters())

vllm/model_executor/models/nemotron.py

@@ -45,7 +45,6 @@ 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 vllm.transformers_utils.configs import NemotronConfig
 
@@ -498,10 +497,8 @@ class NemotronForCausalLM(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)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/nemotron_h.py

@@ -54,7 +54,6 @@ from vllm.model_executor.models.mamba_cache import (MambaCacheManager,
 from vllm.model_executor.models.utils import (
     AutoWeightsLoader, WeightsMapper, make_empty_intermediate_tensors_factory,
     make_layers, maybe_prefix)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.configs import NemotronHConfig
 from vllm.utils import LayerBlockType
@@ -622,10 +621,8 @@ class NemotronHForCausalLM(nn.Module, HasInnerState, 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)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str,

vllm/model_executor/models/nemotron_nas.py

@@ -44,7 +44,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import (
     default_weight_loader, maybe_remap_kv_scale_name)
 from vllm.model_executor.models.llama import LlamaAttention, LlamaMLP
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .interfaces import HasNoOps, SupportsLoRA, SupportsPP
@@ -468,10 +467,8 @@ class DeciLMForCausalLM(nn.Module, SupportsLoRA, SupportsPP, HasNoOps):
     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/nemotron_vl.py

@@ -26,7 +26,6 @@ from vllm.model_executor.models.internvl import (
     BaseInternVLProcessingInfo, InternVLImageEmbeddingInputs,
     InternVLImageInputs, InternVLImagePixelInputs, InternVLProcessor)
 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.image import convert_image_mode
 from vllm.multimodal.inputs import NestedTensors
@@ -632,10 +631,8 @@ class LlamaNemotronVLChatModel(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/olmo.py

@@ -45,7 +45,6 @@ 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
@@ -391,10 +390,8 @@ class OlmoForCausalLM(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/olmo2.py

@@ -54,7 +54,6 @@ from vllm.model_executor.models.interfaces import SupportsLoRA, SupportsPP
 from vllm.model_executor.models.utils import (
     AutoWeightsLoader, extract_layer_index, is_pp_missing_parameter,
     make_empty_intermediate_tensors_factory, make_layers, maybe_prefix)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.configs import Olmo3Config
 
@@ -427,10 +426,8 @@ class Olmo2ForCausalLM(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, torch.Tensor]]):

vllm/model_executor/models/olmoe.py

@@ -41,7 +41,6 @@ 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
@@ -471,10 +470,8 @@ class OlmoeForCausalLM(nn.Module, 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/opt.py

@@ -41,7 +41,6 @@ from vllm.model_executor.layers.quantization import QuantizationConfig
 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
@@ -399,10 +398,8 @@ class OPTForCausalLM(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/orion.py

@@ -28,7 +28,6 @@ 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
@@ -339,10 +338,8 @@ class OrionForCausalLM(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/ovis.py

@@ -39,7 +39,6 @@ from vllm.model_executor.models.siglip import SiglipVisionModel
 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)
@@ -558,9 +557,8 @@ class Ovis(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,