[Deprecation] Remove inputs arg fallback in Engine classes (#18799)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-06-09 10:19:56 +08:00
committed by GitHub
parent 2ffb9b6e07
commit e31ae3de36
4 changed files with 21 additions and 268 deletions
+2 -43
View File
@@ -4,9 +4,7 @@
import uuid
from dataclasses import dataclass, field
from enum import Enum
from typing import List, Mapping, Optional, Union, overload
from typing_extensions import deprecated
from typing import List, Mapping, Optional, Union
from vllm import PoolingParams
from vllm.inputs import PromptType
@@ -14,7 +12,7 @@ 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 Device, deprecate_kwargs
from vllm.utils import Device
VLLM_RPC_SUCCESS_STR = "SUCCESS"
@@ -38,7 +36,6 @@ class RPCProcessRequest:
prompt_adapter_request: Optional[PromptAdapterRequest] = None
priority: int = 0
@overload
def __init__(
self,
prompt: PromptType,
@@ -49,44 +46,6 @@ class RPCProcessRequest:
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
priority: int = 0,
) -> None:
...
@overload
@deprecated("'inputs' will be renamed to 'prompt")
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,
priority: int = 0,
) -> 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,
priority: int = 0,
*,
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
+7 -82
View File
@@ -6,13 +6,12 @@ import copy
import pickle
from contextlib import contextmanager, suppress
from typing import (Any, AsyncGenerator, Dict, Iterator, List, Mapping,
Optional, Union, cast, overload)
Optional, Union, cast)
import cloudpickle
import psutil
import zmq
import zmq.asyncio
from typing_extensions import deprecated
from zmq import Frame # type: ignore[attr-defined]
from zmq.asyncio import Socket
@@ -49,7 +48,7 @@ from vllm.outputs import PoolingRequestOutput, 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 Device, deprecate_kwargs
from vllm.utils import Device
logger = init_logger(__name__)
@@ -442,7 +441,6 @@ class MQLLMEngineClient(EngineClient):
def dead_error(self) -> BaseException:
return ENGINE_DEAD_ERROR(self._errored_with)
@overload
def generate(
self,
prompt: PromptType,
@@ -452,39 +450,6 @@ class MQLLMEngineClient(EngineClient):
trace_headers: Optional[Mapping[str, str]] = None,
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
priority: int = 0,
) -> AsyncGenerator[RequestOutput, None]:
...
@overload
@deprecated("'inputs' will be renamed to 'prompt")
def generate(
self,
*,
inputs: 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,
priority: int = 0,
) -> 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,
priority: int = 0,
*,
inputs: Optional[PromptType] = None # DEPRECATED
) -> AsyncGenerator[RequestOutput, None]:
"""Generate outputs for a request.
@@ -506,16 +471,12 @@ class MQLLMEngineClient(EngineClient):
Any priority other than 0 will lead to an error if the
scheduling policy is not "priority".
"""
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 cast(
AsyncGenerator[RequestOutput, None],
self._process_request(prompt, sampling_params, request_id,
lora_request, trace_headers,
prompt_adapter_request, priority))
return self._process_request(prompt, sampling_params, request_id,
lora_request, trace_headers,
prompt_adapter_request, priority)
@overload
def encode(
self,
prompt: PromptType,
@@ -524,37 +485,6 @@ class MQLLMEngineClient(EngineClient):
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
priority: int = 0,
) -> AsyncGenerator[PoolingRequestOutput, None]:
...
@overload
@deprecated("'inputs' will be renamed to 'prompt")
def encode(
self,
*,
inputs: PromptType,
pooling_params: PoolingParams,
request_id: str,
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
priority: int = 0,
) -> AsyncGenerator[PoolingRequestOutput, 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,
priority: int = 0,
*,
inputs: Optional[PromptType] = None # DEPRECATED
) -> AsyncGenerator[PoolingRequestOutput, None]:
"""Generate outputs for a request from a pooling model.
@@ -575,11 +505,6 @@ class MQLLMEngineClient(EngineClient):
The output `PoolingRequestOutput` 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 cast(
AsyncGenerator[PoolingRequestOutput, None],
self._process_request(prompt,