[V0 Deprecation] Remove vllm.worker and update according imports (#25901)
This commit is contained in:
@@ -19,7 +19,7 @@ from vllm.sequence import ExecuteModelRequest
|
||||
from vllm.tasks import SupportedTask
|
||||
from vllm.utils import make_async
|
||||
from vllm.v1.outputs import PoolerOutput, SamplerOutput
|
||||
from vllm.worker.worker_base import WorkerBase
|
||||
from vllm.v1.worker.worker_base import WorkerBase
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
@@ -30,7 +30,7 @@ class ExecutorBase(ABC):
|
||||
"""Base class for all executors.
|
||||
|
||||
An executor is responsible for executing the model on one device,
|
||||
or it can be a distributed executor
|
||||
or it can be a distributed executor
|
||||
that can execute the model on multiple devices.
|
||||
"""
|
||||
|
||||
@@ -83,7 +83,7 @@ class ExecutorBase(ABC):
|
||||
|
||||
Returns:
|
||||
A list containing the results from each worker.
|
||||
|
||||
|
||||
Note:
|
||||
It is recommended to use this API to only pass control messages,
|
||||
and set up data-plane communication to pass data.
|
||||
@@ -100,7 +100,7 @@ class ExecutorBase(ABC):
|
||||
|
||||
Returns a tuple `(num_gpu_blocks, num_cpu_blocks)`, where
|
||||
`num_gpu_blocks` are blocks that are "active" on the device and can be
|
||||
appended to.
|
||||
appended to.
|
||||
`num_cpu_blocks` refers to "swapped" blocks in CPU memory and cannot be
|
||||
appended to.
|
||||
"""
|
||||
@@ -327,7 +327,7 @@ class DistributedExecutorBase(ExecutorBase):
|
||||
run only in the remote TP workers, not the driver worker.
|
||||
It will also be run asynchronously and return a list of futures
|
||||
rather than blocking on the results.
|
||||
|
||||
|
||||
# TODO: simplify and merge with collective_rpc
|
||||
"""
|
||||
raise NotImplementedError
|
||||
|
||||
@@ -16,7 +16,7 @@ from vllm.logger import init_logger
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sequence import ExecuteModelRequest, IntermediateTensors
|
||||
from vllm.utils import get_ip
|
||||
from vllm.worker.worker_base import WorkerWrapperBase
|
||||
from vllm.v1.worker.worker_base import WorkerWrapperBase
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from vllm.v1.core.sched.output import SchedulerOutput
|
||||
|
||||
@@ -19,7 +19,7 @@ from vllm.utils import (get_distributed_init_method, get_ip, get_open_port,
|
||||
from vllm.v1.engine import ReconfigureDistributedRequest, ReconfigureRankType
|
||||
from vllm.v1.executor.utils import get_and_update_mm_cache
|
||||
from vllm.v1.outputs import AsyncModelRunnerOutput
|
||||
from vllm.worker.worker_base import WorkerWrapperBase
|
||||
from vllm.v1.worker.worker_base import WorkerWrapperBase
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
@@ -160,10 +160,10 @@ class ExecutorWithExternalLauncher(UniProcExecutor):
|
||||
"""
|
||||
Determine the number of available KV blocks.
|
||||
Add an additional all_reduce to get the min across all ranks.
|
||||
Note that even if we have the same `gpu_memory_utilization` and
|
||||
`swap_space`, the available memory in every rank might still
|
||||
differ because NCCL can take different amounts of memory in
|
||||
different ranks. Therefore, it is necessary to test if all ranks
|
||||
Note that even if we have the same `gpu_memory_utilization` and
|
||||
`swap_space`, the available memory in every rank might still
|
||||
differ because NCCL can take different amounts of memory in
|
||||
different ranks. Therefore, it is necessary to test if all ranks
|
||||
agree on the same KV cache configuration.
|
||||
"""
|
||||
a, b = super().determine_num_available_blocks()
|
||||
|
||||
Reference in New Issue
Block a user