This commit is contained in:
@@ -2,8 +2,8 @@ import asyncio
|
||||
import time
|
||||
import weakref
|
||||
from functools import partial
|
||||
from typing import (Any, AsyncGenerator, Callable, Coroutine, Dict, Iterable,
|
||||
List, Mapping, Optional, Set, Tuple, Type, Union, overload)
|
||||
from typing import (Any, AsyncGenerator, Callable, Dict, Iterable, List,
|
||||
Mapping, Optional, Set, Tuple, Type, Union)
|
||||
from weakref import ReferenceType
|
||||
|
||||
import vllm.envs as envs
|
||||
@@ -17,7 +17,7 @@ from vllm.engine.metrics_types import StatLoggerBase
|
||||
from vllm.executor.executor_base import ExecutorAsyncBase
|
||||
from vllm.executor.gpu_executor import GPUExecutorAsync
|
||||
from vllm.executor.ray_utils import initialize_ray_cluster
|
||||
from vllm.inputs import PromptType
|
||||
from vllm.inputs import PromptInputs
|
||||
from vllm.logger import init_logger
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.model_executor.layers.sampler import SamplerOutput
|
||||
@@ -28,7 +28,7 @@ from vllm.sampling_params import SamplingParams
|
||||
from vllm.sequence import ExecuteModelRequest
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
from vllm.usage.usage_lib import UsageContext
|
||||
from vllm.utils import deprecate_kwargs, weak_bind
|
||||
from vllm.utils import weak_bind
|
||||
|
||||
logger = init_logger(__name__)
|
||||
ENGINE_ITERATION_TIMEOUT_S = envs.VLLM_ENGINE_ITERATION_TIMEOUT_S
|
||||
@@ -402,54 +402,17 @@ class _AsyncLLMEngine(LLMEngine):
|
||||
"""Stop the remote worker execution loop."""
|
||||
await self.model_executor.stop_remote_worker_execution_loop_async()
|
||||
|
||||
@overload # DEPRECATED
|
||||
async def add_request_async(
|
||||
self,
|
||||
request_id: str,
|
||||
*,
|
||||
inputs: PromptType,
|
||||
inputs: PromptInputs,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> None:
|
||||
...
|
||||
|
||||
@overload
|
||||
async def add_request_async(
|
||||
self,
|
||||
request_id: str,
|
||||
prompt: PromptType,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> None:
|
||||
...
|
||||
|
||||
@deprecate_kwargs(
|
||||
"inputs",
|
||||
additional_message="Please use the 'prompt' parameter instead.",
|
||||
)
|
||||
async def add_request_async(
|
||||
self,
|
||||
request_id: str,
|
||||
prompt: Optional[PromptType] = None,
|
||||
params: Optional[Union[SamplingParams, PoolingParams]] = None,
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
*,
|
||||
inputs: Optional[PromptType] = None, # DEPRECATED
|
||||
) -> None:
|
||||
"""Async version of :meth:`add_request`."""
|
||||
if inputs is not None:
|
||||
prompt = inputs
|
||||
assert prompt is not None and params is not None
|
||||
|
||||
if lora_request is not None and not self.lora_config:
|
||||
raise ValueError(f"Got lora_request {lora_request} but LoRA is "
|
||||
"not enabled!")
|
||||
@@ -457,7 +420,7 @@ class _AsyncLLMEngine(LLMEngine):
|
||||
arrival_time = time.time()
|
||||
|
||||
preprocessed_inputs = await self.input_preprocessor.preprocess_async(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
prompt_adapter_request=prompt_adapter_request,
|
||||
@@ -811,55 +774,16 @@ class AsyncLLMEngine:
|
||||
|
||||
# This method does not need to be async, but kept that way
|
||||
# for backwards compatibility.
|
||||
@overload # DEPRECATED
|
||||
def add_request(
|
||||
self,
|
||||
request_id: str,
|
||||
*,
|
||||
inputs: PromptType,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> Coroutine[None, None, AsyncGenerator[Union[
|
||||
RequestOutput, EmbeddingRequestOutput], None]]:
|
||||
...
|
||||
|
||||
@overload
|
||||
def add_request(
|
||||
self,
|
||||
request_id: str,
|
||||
prompt: PromptType,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> Coroutine[None, None, AsyncGenerator[Union[
|
||||
RequestOutput, EmbeddingRequestOutput], None]]:
|
||||
...
|
||||
|
||||
@deprecate_kwargs(
|
||||
"inputs",
|
||||
additional_message="Please use the 'prompt' parameter instead.",
|
||||
)
|
||||
async def add_request(
|
||||
self,
|
||||
request_id: str,
|
||||
prompt: Optional[PromptType] = None,
|
||||
params: Optional[Union[SamplingParams, PoolingParams]] = None,
|
||||
inputs: PromptInputs,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
*,
|
||||
inputs: Optional[PromptType] = None, # DEPRECATED
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None
|
||||
) -> AsyncGenerator[Union[RequestOutput, EmbeddingRequestOutput], None]:
|
||||
if inputs is not None:
|
||||
prompt = inputs
|
||||
assert prompt is not None and params is not None
|
||||
|
||||
if not self.is_running:
|
||||
if self.start_engine_loop:
|
||||
self.start_background_loop()
|
||||
@@ -873,7 +797,7 @@ class AsyncLLMEngine:
|
||||
stream = self._request_tracker.add_request(
|
||||
request_id,
|
||||
verbose=self.log_requests,
|
||||
prompt=prompt,
|
||||
inputs=inputs,
|
||||
params=params,
|
||||
arrival_time=arrival_time or time.time(),
|
||||
lora_request=lora_request,
|
||||
@@ -884,7 +808,7 @@ class AsyncLLMEngine:
|
||||
|
||||
async def generate(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
sampling_params: SamplingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
@@ -898,7 +822,8 @@ class AsyncLLMEngine:
|
||||
from the LLMEngine to the caller.
|
||||
|
||||
Args:
|
||||
prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
|
||||
inputs: The inputs to the LLM. See
|
||||
:class:`~vllm.inputs.PromptInputs`
|
||||
for more details about the format of each input.
|
||||
sampling_params: The sampling parameters of the request.
|
||||
request_id: The unique id of the request.
|
||||
@@ -956,7 +881,7 @@ class AsyncLLMEngine:
|
||||
"""
|
||||
async for output in await self.add_request(
|
||||
request_id,
|
||||
prompt,
|
||||
inputs,
|
||||
sampling_params,
|
||||
lora_request=lora_request,
|
||||
trace_headers=trace_headers,
|
||||
@@ -966,7 +891,7 @@ class AsyncLLMEngine:
|
||||
|
||||
async def encode(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
pooling_params: PoolingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
@@ -979,7 +904,8 @@ class AsyncLLMEngine:
|
||||
from the LLMEngine to the caller.
|
||||
|
||||
Args:
|
||||
prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
|
||||
inputs: The inputs to the LLM. See
|
||||
:class:`~vllm.inputs.PromptInputs`
|
||||
for more details about the format of each input.
|
||||
pooling_params: The pooling parameters of the request.
|
||||
request_id: The unique id of the request.
|
||||
@@ -1033,7 +959,7 @@ class AsyncLLMEngine:
|
||||
"""
|
||||
async for output in await self.add_request(
|
||||
request_id,
|
||||
prompt,
|
||||
inputs,
|
||||
pooling_params,
|
||||
lora_request=lora_request,
|
||||
trace_headers=trace_headers,
|
||||
|
||||
@@ -6,7 +6,7 @@ from functools import partial
|
||||
from typing import (TYPE_CHECKING, Any, Callable, ClassVar, Deque, Dict,
|
||||
Iterable, List, Mapping, NamedTuple, Optional)
|
||||
from typing import Sequence as GenericSequence
|
||||
from typing import Set, Type, Union, overload
|
||||
from typing import Set, Type, Union
|
||||
|
||||
import torch
|
||||
from typing_extensions import TypeVar
|
||||
@@ -29,7 +29,7 @@ from vllm.executor.executor_base import ExecutorBase
|
||||
from vllm.executor.gpu_executor import GPUExecutor
|
||||
from vllm.executor.ray_utils import initialize_ray_cluster
|
||||
from vllm.inputs import (INPUT_REGISTRY, EncoderDecoderLLMInputs,
|
||||
InputRegistry, LLMInputs, PromptType)
|
||||
InputRegistry, LLMInputs, PromptInputs)
|
||||
from vllm.inputs.preprocess import InputPreprocessor
|
||||
from vllm.logger import init_logger
|
||||
from vllm.lora.request import LoRARequest
|
||||
@@ -51,7 +51,7 @@ from vllm.transformers_utils.tokenizer_group import (
|
||||
BaseTokenizerGroup, init_tokenizer_from_configs)
|
||||
from vllm.usage.usage_lib import (UsageContext, is_usage_stats_enabled,
|
||||
usage_message)
|
||||
from vllm.utils import Counter, Device, deprecate_kwargs, weak_bind
|
||||
from vllm.utils import Counter, Device, weak_bind
|
||||
from vllm.version import __version__ as VLLM_VERSION
|
||||
|
||||
logger = init_logger(__name__)
|
||||
@@ -689,51 +689,16 @@ class LLMEngine:
|
||||
def stop_remote_worker_execution_loop(self) -> None:
|
||||
self.model_executor.stop_remote_worker_execution_loop()
|
||||
|
||||
@overload # DEPRECATED
|
||||
def add_request(
|
||||
self,
|
||||
request_id: str,
|
||||
*,
|
||||
inputs: PromptType,
|
||||
inputs: PromptInputs,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
priority: int = 0,
|
||||
) -> None:
|
||||
...
|
||||
|
||||
@overload
|
||||
def add_request(
|
||||
self,
|
||||
request_id: str,
|
||||
prompt: PromptType,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
priority: int = 0,
|
||||
) -> None:
|
||||
...
|
||||
|
||||
@deprecate_kwargs(
|
||||
"inputs",
|
||||
additional_message="Please use the 'prompt' parameter instead.",
|
||||
)
|
||||
def add_request(
|
||||
self,
|
||||
request_id: str,
|
||||
prompt: Optional[PromptType] = None,
|
||||
params: Optional[Union[SamplingParams, PoolingParams]] = None,
|
||||
arrival_time: Optional[float] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
priority: int = 0,
|
||||
*,
|
||||
inputs: Optional[PromptType] = None, # DEPRECATED
|
||||
) -> None:
|
||||
"""Add a request to the engine's request pool.
|
||||
|
||||
@@ -743,7 +708,8 @@ class LLMEngine:
|
||||
|
||||
Args:
|
||||
request_id: The unique ID of the request.
|
||||
prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
|
||||
inputs: The inputs to the LLM. See
|
||||
:class:`~vllm.inputs.PromptInputs`
|
||||
for more details about the format of each input.
|
||||
params: Parameters for sampling or pooling.
|
||||
:class:`~vllm.SamplingParams` for text generation.
|
||||
@@ -778,10 +744,6 @@ class LLMEngine:
|
||||
>>> # continue the request processing
|
||||
>>> ...
|
||||
"""
|
||||
if inputs is not None:
|
||||
prompt = inputs
|
||||
assert prompt is not None and params is not None
|
||||
|
||||
if lora_request is not None and not self.lora_config:
|
||||
raise ValueError(f"Got lora_request {lora_request} but LoRA is "
|
||||
"not enabled!")
|
||||
@@ -794,7 +756,7 @@ class LLMEngine:
|
||||
arrival_time = time.time()
|
||||
|
||||
preprocessed_inputs = self.input_preprocessor.preprocess(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
prompt_adapter_request=prompt_adapter_request,
|
||||
|
||||
@@ -1,14 +1,13 @@
|
||||
from dataclasses import dataclass
|
||||
from enum import Enum
|
||||
from typing import List, Mapping, Optional, Union, overload
|
||||
from typing import List, Mapping, Optional, Union
|
||||
|
||||
from vllm import PoolingParams
|
||||
from vllm.inputs import PromptType
|
||||
from vllm.inputs import PromptInputs
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.outputs import RequestOutput
|
||||
from vllm.prompt_adapter.request import PromptAdapterRequest
|
||||
from vllm.sampling_params import SamplingParams
|
||||
from vllm.utils import deprecate_kwargs
|
||||
|
||||
VLLM_RPC_SUCCESS_STR = "SUCCESS"
|
||||
|
||||
@@ -24,67 +23,13 @@ class MQEngineDeadError(RuntimeError):
|
||||
|
||||
@dataclass
|
||||
class RPCProcessRequest:
|
||||
prompt: PromptType
|
||||
inputs: PromptInputs
|
||||
params: Union[SamplingParams, PoolingParams]
|
||||
request_id: str
|
||||
lora_request: Optional[LoRARequest] = None
|
||||
trace_headers: Optional[Mapping[str, str]] = None
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None
|
||||
|
||||
@overload # DEPRECATED
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
inputs: PromptType,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> None:
|
||||
...
|
||||
|
||||
@overload
|
||||
def __init__(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> None:
|
||||
...
|
||||
|
||||
@deprecate_kwargs(
|
||||
"inputs",
|
||||
additional_message="Please use the 'prompt' parameter instead.",
|
||||
)
|
||||
def __init__(
|
||||
self,
|
||||
prompt: Optional[PromptType] = None,
|
||||
params: Optional[Union[SamplingParams, PoolingParams]] = None,
|
||||
request_id: Optional[str] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
*,
|
||||
inputs: Optional[PromptType] = None, # DEPRECATED
|
||||
) -> None:
|
||||
if inputs is not None:
|
||||
prompt = inputs
|
||||
assert (prompt is not None and params is not None
|
||||
and request_id is not None)
|
||||
|
||||
super().__init__()
|
||||
|
||||
self.prompt = prompt
|
||||
self.params = params
|
||||
self.request_id = request_id
|
||||
self.lora_request = lora_request
|
||||
self.trace_headers = trace_headers
|
||||
self.prompt_adapter_request = prompt_adapter_request
|
||||
|
||||
|
||||
@dataclass
|
||||
class RPCError:
|
||||
|
||||
@@ -3,7 +3,7 @@ import copy
|
||||
import pickle
|
||||
from contextlib import contextmanager, suppress
|
||||
from typing import (Any, AsyncGenerator, Dict, Iterator, Mapping, Optional,
|
||||
Union, overload)
|
||||
Union)
|
||||
|
||||
import cloudpickle
|
||||
import zmq
|
||||
@@ -25,14 +25,13 @@ from vllm.engine.multiprocessing import (ENGINE_DEAD_ERROR, IPC_DATA_EXT,
|
||||
RPCUProfileRequest)
|
||||
# yapf: enable
|
||||
from vllm.envs import VLLM_RPC_TIMEOUT
|
||||
from vllm.inputs import PromptType
|
||||
from vllm.inputs import PromptInputs
|
||||
from vllm.logger import init_logger
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.outputs import EmbeddingRequestOutput, RequestOutput
|
||||
from vllm.prompt_adapter.request import PromptAdapterRequest
|
||||
from vllm.sampling_params import SamplingParams
|
||||
from vllm.transformers_utils.tokenizer_group import init_tokenizer_from_configs
|
||||
from vllm.utils import deprecate_kwargs
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
@@ -368,45 +367,14 @@ class MQLLMEngineClient:
|
||||
def dead_error(self) -> BaseException:
|
||||
return ENGINE_DEAD_ERROR(self._errored_with)
|
||||
|
||||
@overload # DEPRECATED
|
||||
def generate(
|
||||
self,
|
||||
*,
|
||||
inputs: PromptType,
|
||||
inputs: PromptInputs,
|
||||
sampling_params: SamplingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> AsyncGenerator[RequestOutput, None]:
|
||||
...
|
||||
|
||||
@overload
|
||||
def generate(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
sampling_params: SamplingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> AsyncGenerator[RequestOutput, None]:
|
||||
...
|
||||
|
||||
@deprecate_kwargs(
|
||||
"inputs",
|
||||
additional_message="Please use the 'prompt' parameter instead.",
|
||||
)
|
||||
def generate(
|
||||
self,
|
||||
prompt: Optional[PromptType] = None,
|
||||
sampling_params: Optional[SamplingParams] = None,
|
||||
request_id: Optional[str] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
*,
|
||||
inputs: Optional[PromptType] = None # DEPRECATED
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None
|
||||
) -> AsyncGenerator[RequestOutput, None]:
|
||||
"""Generate outputs for a request.
|
||||
|
||||
@@ -415,7 +383,8 @@ class MQLLMEngineClient:
|
||||
from the LLMEngine to the caller.
|
||||
|
||||
Args:
|
||||
prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
|
||||
inputs: The inputs to the LLM. See
|
||||
:class:`~vllm.inputs.PromptInputs`
|
||||
for more details about the format of each input.
|
||||
sampling_params: The sampling parameters of the request.
|
||||
request_id: The unique id of the request.
|
||||
@@ -424,51 +393,17 @@ class MQLLMEngineClient:
|
||||
prompt_adapter_request: Prompt Adapter request to use
|
||||
for generation, if any.
|
||||
"""
|
||||
if inputs is not None:
|
||||
prompt = inputs
|
||||
assert (prompt is not None and sampling_params is not None
|
||||
and request_id is not None)
|
||||
|
||||
return self._process_request(prompt, sampling_params, request_id,
|
||||
return self._process_request(inputs, sampling_params, request_id,
|
||||
lora_request, trace_headers,
|
||||
prompt_adapter_request)
|
||||
|
||||
@overload # DEPRECATED
|
||||
def encode(
|
||||
self,
|
||||
*,
|
||||
inputs: PromptType,
|
||||
inputs: PromptInputs,
|
||||
pooling_params: PoolingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
|
||||
...
|
||||
|
||||
@overload
|
||||
def encode(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
pooling_params: PoolingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
|
||||
...
|
||||
|
||||
@deprecate_kwargs(
|
||||
"inputs",
|
||||
additional_message="Please use the 'prompt' parameter instead.",
|
||||
)
|
||||
def encode(
|
||||
self,
|
||||
prompt: Optional[PromptType] = None,
|
||||
pooling_params: Optional[PoolingParams] = None,
|
||||
request_id: Optional[str] = None,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
*,
|
||||
inputs: Optional[PromptType] = None # DEPRECATED
|
||||
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
|
||||
"""Generate outputs for a request from an embedding model.
|
||||
|
||||
@@ -477,7 +412,8 @@ class MQLLMEngineClient:
|
||||
from the LLMEngine to the caller.
|
||||
|
||||
Args:
|
||||
prompt: The prompt to the LLM. See :class:`~vllm.inputs.PromptType`
|
||||
inputs: The inputs to the LLM. See
|
||||
:class:`~vllm.inputs.PromptInputs`
|
||||
for more details about the format of each input.
|
||||
pooling_params: The pooling parameters of the request.
|
||||
request_id: The unique id of the request.
|
||||
@@ -488,17 +424,12 @@ class MQLLMEngineClient:
|
||||
The output `EmbeddingRequestOutput` objects from the LLMEngine
|
||||
for the request.
|
||||
"""
|
||||
if inputs is not None:
|
||||
prompt = inputs
|
||||
assert (prompt is not None and pooling_params is not None
|
||||
and request_id is not None)
|
||||
|
||||
return self._process_request(prompt, pooling_params, request_id,
|
||||
return self._process_request(inputs, pooling_params, request_id,
|
||||
lora_request, trace_headers)
|
||||
|
||||
async def _process_request(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
params: Union[SamplingParams, PoolingParams],
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
@@ -531,7 +462,7 @@ class MQLLMEngineClient:
|
||||
|
||||
request_bytes = pickle.dumps(
|
||||
RPCProcessRequest(
|
||||
prompt=prompt,
|
||||
inputs=inputs,
|
||||
params=params,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
|
||||
@@ -278,7 +278,7 @@ class MQLLMEngine:
|
||||
try:
|
||||
self.engine.add_request(
|
||||
request_id=request_id,
|
||||
prompt=request.prompt,
|
||||
inputs=request.inputs,
|
||||
params=request.params,
|
||||
lora_request=request.lora_request,
|
||||
trace_headers=request.trace_headers,
|
||||
|
||||
@@ -3,7 +3,7 @@ from typing import (AsyncGenerator, List, Mapping, Optional, Protocol,
|
||||
|
||||
from vllm.config import DecodingConfig, ModelConfig
|
||||
from vllm.core.scheduler import SchedulerOutputs
|
||||
from vllm.inputs.data import PromptType
|
||||
from vllm.inputs.data import PromptInputs
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.model_executor.layers.sampler import SamplerOutput
|
||||
from vllm.outputs import EmbeddingRequestOutput, RequestOutput
|
||||
@@ -35,19 +35,19 @@ class EngineClient(Protocol):
|
||||
|
||||
def generate(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
sampling_params: SamplingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None
|
||||
) -> AsyncGenerator[RequestOutput, None]:
|
||||
"""Generate outputs for a request."""
|
||||
"""Generates outputs for a request"""
|
||||
...
|
||||
|
||||
def encode(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
pooling_params: PoolingParams,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
|
||||
Reference in New Issue
Block a user