[Core] Logprobs support in Multi-step (#7652)

vllm/executor/cpu_executor.py

@@ -11,8 +11,9 @@ from vllm.executor.multiproc_worker_utils import (ProcessWorkerWrapper,
                                                   ResultHandler, WorkerMonitor)
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
+from vllm.model_executor.layers.sampler import SamplerOutput
 from vllm.prompt_adapter.request import PromptAdapterRequest
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 from vllm.utils import (GiB_bytes, get_distributed_init_method, get_open_port,
                         get_vllm_instance_id, make_async)
 from vllm.worker.worker_base import WorkerWrapperBase

vllm/executor/distributed_gpu_executor.py

@@ -6,7 +6,8 @@ from vllm.executor.executor_base import ExecutorAsyncBase
 from vllm.executor.gpu_executor import GPUExecutor
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 
 logger = init_logger(__name__)
 

vllm/executor/executor_base.py

@@ -6,8 +6,9 @@ from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
                          PromptAdapterConfig, SchedulerConfig,
                          SpeculativeConfig)
 from vllm.lora.request import LoRARequest
+from vllm.model_executor.layers.sampler import SamplerOutput
 from vllm.prompt_adapter.request import PromptAdapterRequest
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 
 
 class ExecutorBase(ABC):

vllm/executor/gpu_executor.py

@@ -3,8 +3,9 @@ from typing import Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
 from vllm.executor.executor_base import ExecutorAsyncBase, ExecutorBase
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
+from vllm.model_executor.layers.sampler import SamplerOutput
 from vllm.prompt_adapter.request import PromptAdapterRequest
-from vllm.sequence import ExecuteModelRequest, PoolerOutput, SamplerOutput
+from vllm.sequence import ExecuteModelRequest, PoolerOutput
 from vllm.utils import (get_distributed_init_method, get_ip, get_open_port,
                         make_async)
 from vllm.worker.worker_base import WorkerBase, WorkerWrapperBase

vllm/executor/multiproc_gpu_executor.py

@@ -14,7 +14,8 @@ from vllm.executor.gpu_executor import create_worker
 from vllm.executor.multiproc_worker_utils import (ProcessWorkerWrapper,
                                                   ResultHandler, WorkerMonitor)
 from vllm.logger import init_logger
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 from vllm.triton_utils import maybe_set_triton_cache_manager
 from vllm.utils import (_run_task_with_lock, cuda_device_count_stateless,
                         get_distributed_init_method, get_open_port,

vllm/executor/neuron_executor.py

@@ -3,7 +3,8 @@ from typing import List, Set, Tuple
 from vllm.executor.executor_base import ExecutorAsyncBase, ExecutorBase
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 from vllm.utils import (get_distributed_init_method, get_ip, get_open_port,
                         make_async)
 

vllm/executor/openvino_executor.py

@@ -9,7 +9,8 @@ from vllm.config import CacheConfig, ModelConfig
 from vllm.executor.executor_base import ExecutorAsyncBase, ExecutorBase
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 from vllm.utils import (GiB_bytes, get_distributed_init_method, get_ip,
                         get_open_port, make_async)
 

vllm/executor/ray_gpu_executor.py

@@ -12,7 +12,8 @@ from vllm.executor.distributed_gpu_executor import (  # yapf: disable
 from vllm.executor.msgspec_utils import encode_hook
 from vllm.executor.ray_utils import RayWorkerWrapper, ray
 from vllm.logger import init_logger
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 from vllm.utils import (_run_task_with_lock, get_distributed_init_method,
                         get_ip, get_open_port, get_vllm_instance_id,
                         make_async)

vllm/executor/ray_tpu_executor.py

@@ -10,7 +10,8 @@ from vllm.executor.executor_base import ExecutorAsyncBase
 from vllm.executor.ray_utils import RayWorkerWrapper, ray
 from vllm.executor.tpu_executor import TPUExecutor
 from vllm.logger import init_logger
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 from vllm.utils import (get_distributed_init_method, get_ip, get_open_port,
                         get_vllm_instance_id, make_async)
 

vllm/executor/tpu_executor.py

@@ -5,7 +5,8 @@ import torch
 from vllm.executor.executor_base import ExecutorAsyncBase, ExecutorBase
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
-from vllm.sequence import ExecuteModelRequest, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest
 from vllm.utils import (get_distributed_init_method, get_ip, get_open_port,
                         make_async)
 

vllm/executor/xpu_executor.py

@@ -9,7 +9,8 @@ from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
 from vllm.executor.executor_base import ExecutorAsyncBase
 from vllm.executor.gpu_executor import GPUExecutor
 from vllm.logger import init_logger
-from vllm.sequence import ExecuteModelRequest, PoolerOutput, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput
+from vllm.sequence import ExecuteModelRequest, PoolerOutput
 from vllm.utils import make_async
 from vllm.worker.worker_base import WorkerBase