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

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

vllm/model_executor/models/interns1.py

@@ -21,7 +21,6 @@ from vllm.config import VllmConfig
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.models.interns1_vit import InternS1VisionModel
 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)
@@ -812,10 +811,8 @@ class InternS1ForConditionalGeneration(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/internvl.py

@@ -25,7 +25,6 @@ from vllm.model_executor.layers.quantization.awq import AWQConfig
 from vllm.model_executor.models.intern_vit import (InternVisionModel,
                                                    InternVisionPatchModel)
 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 (MultiModalDataDict, MultiModalFieldConfig,
@@ -1399,10 +1398,8 @@ class InternVLChatModel(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/jais.py

@@ -42,7 +42,6 @@ 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 vllm.transformers_utils.configs import JAISConfig
 
@@ -332,10 +331,8 @@ class JAISLMHeadModel(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/jamba.py

@@ -32,7 +32,6 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.models.llama import LlamaMLP as JambaMLP
 from vllm.model_executor.models.mamba_cache import (MambaCacheManager,
                                                     MambaCacheParams)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 from vllm.utils import LayerBlockType
 
@@ -581,10 +580,8 @@ class JambaForCausalLM(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/keye.py

@@ -21,7 +21,6 @@ from vllm.attention.layer import check_upstream_fa_availability
 from vllm.config import VllmConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.logger import init_logger
-from vllm.model_executor import SamplingMetadata
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                QKVParallelLinear,
                                                RowParallelLinear)
@@ -1556,10 +1555,8 @@ class BaseKeyeModule(nn.Module):
     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/kimi_vl.py

@@ -67,7 +67,6 @@ from vllm.model_executor.models.interfaces import (SupportsMultiModal,
                                                    SupportsPP)
 from vllm.model_executor.models.moonvit import MoonVitPretrainedModel
 from vllm.model_executor.models.utils import merge_multimodal_embeddings
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems, NestedTensors)
@@ -484,10 +483,8 @@ class KimiVLForConditionalGeneration(nn.Module, SupportsMultiModal,
         return hidden_states
 
     def compute_logits(self, hidden_states: torch.Tensor,
-                       sampling_metadata: SamplingMetadata,
                        **kwargs) -> torch.Tensor:
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata, **kwargs)
+        logits = self.logits_processor(self.lm_head, hidden_states, **kwargs)
         return logits
 
     def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):

vllm/model_executor/models/lfm2.py

@@ -27,7 +27,6 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 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 (HasInnerState, IsHybrid, SupportsLoRA, SupportsPP,
@@ -542,10 +541,8 @@ class Lfm2ForCausalLM(nn.Module, HasInnerState, 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/llama.py

@@ -48,7 +48,6 @@ 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 SupportsEagle3, SupportsLoRA, SupportsPP
@@ -601,10 +600,8 @@ class LlamaForCausalLM(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/llama_eagle3.py

@@ -21,7 +21,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.models.llama import (LlamaDecoderLayer,
                                               LlamaForCausalLM)
-from vllm.v1.sample.metadata import SamplingMetadata
 
 from .utils import AutoWeightsLoader, maybe_prefix
 
@@ -244,10 +243,8 @@ class Eagle3LlamaForCausalLM(LlamaForCausalLM):
     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)
         if self.draft_id_to_target_id is None:
             assert logits.shape[1] == self.config.vocab_size, \
                 "Expected logits to have shape " \

vllm/model_executor/models/llava.py

@@ -20,7 +20,6 @@ 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.cache import BaseMultiModalProcessorCache
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
@@ -760,10 +759,8 @@ class LlavaForConditionalGeneration(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/llava_next.py

@@ -13,7 +13,6 @@ from transformers.models.llava_next.modeling_llava_next import (
     get_anyres_image_grid_shape, unpad_image)
 
 from vllm.config import VllmConfig
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import MultiModalFieldConfig
 from vllm.multimodal.parse import ImageSize
@@ -563,10 +562,8 @@ model_executor.models.llava_next.LlavaNextProcessingInfo.get_num_image_tokens].
     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/llava_next_video.py

@@ -13,7 +13,6 @@ from transformers import (BatchFeature, LlavaNextVideoConfig,
 from vllm.config import VllmConfig
 from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.models.clip import CLIPVisionModel
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems)
@@ -464,10 +463,8 @@ class LlavaNextVideoForConditionalGeneration(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/llava_onevision.py

@@ -14,7 +14,6 @@ from transformers.models.llava_onevision.modeling_llava_onevision import (
 
 from vllm.config import VllmConfig
 from vllm.model_executor.layers.activation import get_act_fn
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems)
@@ -934,10 +933,8 @@ class LlavaOnevisionForConditionalGeneration(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/mamba.py

@@ -26,7 +26,6 @@ from vllm.model_executor.models.interfaces import (HasInnerState,
                                                    IsAttentionFree, SupportsPP)
 from vllm.model_executor.models.mamba_cache import (MambaCacheManager,
                                                     MambaCacheParams)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 from vllm.utils import LayerBlockType
 
@@ -299,10 +298,8 @@ class MambaForCausalLM(nn.Module, HasInnerState, IsAttentionFree, SupportsPP):
     def get_seqlen_agnostic_capture_inputs(self, batch_size: int):
         return self.mamba_cache.get_seqlen_agnostic_capture_inputs(batch_size)
 
-    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/mamba2.py

@@ -30,7 +30,6 @@ from vllm.model_executor.models.interfaces import (HasInnerState,
                                                    IsAttentionFree)
 from vllm.model_executor.models.mamba_cache import (MambaCacheManager,
                                                     MambaCacheParams)
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 from vllm.utils import LayerBlockType
 
@@ -335,10 +334,8 @@ class Mamba2ForCausalLM(nn.Module, HasInnerState, IsAttentionFree):
     def get_seqlen_agnostic_capture_inputs(self, batch_size: int):
         return self.mamba_cache.get_seqlen_agnostic_capture_inputs(batch_size)
 
-    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/medusa.py

@@ -104,12 +104,11 @@ class Medusa(nn.Module):
     def compute_logits(
         self,
         hidden_states: list[torch.Tensor],
-        sampling_metadata,
     ) -> list[torch.Tensor]:
         logits_lst: list[torch.Tensor] = []
 
         for hs, lm_head in zip(hidden_states, self.lm_heads):
-            _logits = self.logits_processor(lm_head, hs, sampling_metadata)
+            _logits = self.logits_processor(lm_head, hs)
 
             if _logits is None:
                 # _logits should only be None on rank > 0, in which case

vllm/model_executor/models/midashenglm.py

@@ -42,7 +42,6 @@ from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.model_loader.utils import set_default_torch_dtype
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
                                     MultiModalKwargsItems)
@@ -784,9 +783,8 @@ class MiDashengLMModel(nn.Module, SupportsMultiModal, SupportsPP):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
-        return self.decoder.compute_logits(hidden_states, sampling_metadata)
+        return self.decoder.compute_logits(hidden_states)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/mimo.py

@@ -41,7 +41,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead
 from vllm.model_executor.model_loader.weight_utils import (
     default_weight_loader, maybe_remap_kv_scale_name)
 from vllm.model_executor.models.qwen2 import Qwen2ForCausalLM, Qwen2Model
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .utils import PPMissingLayer, is_pp_missing_parameter, maybe_prefix
@@ -183,9 +182,7 @@ class MiMoForCausalLM(Qwen2ForCausalLM, nn.Module):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
     ) -> Optional[torch.Tensor]:
         hidden_states = self.model.norm(hidden_states)
-        logits = self.logits_processor(self.lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(self.lm_head, hidden_states)
         return logits

vllm/model_executor/models/mimo_mtp.py

@@ -34,7 +34,6 @@ 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.models.qwen2 import Qwen2DecoderLayer
-from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
 from .utils import maybe_prefix
@@ -140,12 +139,10 @@ class MiMoMultiTokenPredictor(nn.Module):
         self,
         hidden_states: torch.Tensor,
         lm_head: ParallelLMHead,
-        sampling_metadata: SamplingMetadata,
         spec_step_idx: int = 0,
     ) -> torch.Tensor:
         self.mtp_layers[str(self.mtp_start_layer_idx + spec_step_idx)]
-        logits = self.logits_processor(lm_head, hidden_states,
-                                       sampling_metadata)
+        logits = self.logits_processor(lm_head, hidden_states)
         return logits
 
 
@@ -178,11 +175,10 @@ class MiMoMTP(nn.Module):
     def compute_logits(
         self,
         hidden_states: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
         spec_step_idx: int = 0,
     ) -> Optional[torch.Tensor]:
         return self.model.compute_logits(hidden_states, self.lm_head,
-                                         sampling_metadata, spec_step_idx)
+                                         spec_step_idx)
 
     def load_weights(self, weights: Iterable[tuple[str,
                                                    torch.Tensor]]) -> set[str]:

vllm/model_executor/models/minicpm.py

@@ -51,7 +51,6 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 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.model_executor.utils import set_weight_attrs
 from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
@@ -583,10 +582,8 @@ class MiniCPMForCausalLM(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,