[Deprecation] Remove scheduled deprecated items (#60923)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-10-10 15:45:33 -04:00
committed by GitHub
parent 187a0eb98a
commit 276fbcff27
7 changed files with 3 additions and 81 deletions
-4
View File
@@ -39,10 +39,6 @@ fn invalid_query(error: QueryRejection) -> ApiError {
ApiError::invalid_request(error.body_text(), Some("mode")) ApiError::invalid_request(error.body_text(), Some("mode"))
} }
// TODO: the Python frontend also accepts the deprecated
// `wait_for_inflight_requests` flag (equivalent to `mode="wait"`); it is
// intentionally omitted here in favor of the `mode` parameter.
/// Pause the scheduler so generation can be halted (e.g. for weight updates). /// Pause the scheduler so generation can be halted (e.g. for weight updates).
pub async fn pause( pub async fn pause(
State(state): State<Arc<AppState>>, State(state): State<Arc<AppState>>,
+2 -23
View File
@@ -625,15 +625,7 @@ def assign_loras(requests, args):
def validate_args(args): def validate_args(args):
"""Validate command-line arguments.""" """Validate command-line arguments."""
# === Deprecation and Defaulting === # === Defaulting ===
if args.dataset is not None:
warnings.warn(
"The '--dataset' argument will be deprecated in the next release. "
"Please use '--dataset-name' and '--dataset-path' instead.",
stacklevel=2,
)
args.dataset_path = args.dataset
if not getattr(args, "tokenizer", None): if not getattr(args, "tokenizer", None):
args.tokenizer = args.model args.tokenizer = args.model
@@ -647,11 +639,7 @@ def validate_args(args):
raise ValueError("--prequeue-requests is not supported with --async-engine") raise ValueError("--prequeue-requests is not supported with --async-engine")
# === Dataset Configuration === # === Dataset Configuration ===
if ( if not args.dataset_path and args.dataset_name != "prefix_repetition":
not args.dataset
and not args.dataset_path
and args.dataset_name not in {"prefix_repetition"}
):
print("When dataset path is not set, it will default to random dataset") print("When dataset path is not set, it will default to random dataset")
args.dataset_name = "random" args.dataset_name = "random"
random_input_len = getattr(args, "random_input_len", None) random_input_len = getattr(args, "random_input_len", None)
@@ -809,15 +797,6 @@ def add_cli_args(parser: FlexibleArgumentParser):
help="Name of the dataset to benchmark on.", help="Name of the dataset to benchmark on.",
default="sharegpt", default="sharegpt",
) )
parser.add_argument(
"--dataset",
type=str,
default=None,
help="Path to the ShareGPT dataset, will be deprecated in\
the next release. The dataset is expected to "
"be a json in form of list[dict[..., conversations: "
"list[dict[..., value: <prompt_or_response>]]]]",
)
parser.add_argument( parser.add_argument(
"--dataset-path", type=str, default=None, help="Path to the dataset" "--dataset-path", type=str, default=None, help="Path to the dataset"
) )
-2
View File
@@ -224,7 +224,6 @@ class EngineClient(ABC):
self, self,
*, *,
mode: "PauseMode" = "abort", mode: "PauseMode" = "abort",
wait_for_inflight_requests: bool = False,
clear_cache: bool = True, clear_cache: bool = True,
) -> None: ) -> None:
"""Pause new generation/encoding requests. """Pause new generation/encoding requests.
@@ -236,7 +235,6 @@ class EngineClient(ABC):
- ``"wait"``: Wait for in-flight requests to complete. - ``"wait"``: Wait for in-flight requests to complete.
- ``"keep"``: Freeze requests in queue; they resume on - ``"keep"``: Freeze requests in queue; they resume on
:meth:`resume_generation`. :meth:`resume_generation`.
wait_for_inflight_requests: DEPRECATED. Use ``mode="wait"`` instead.
clear_cache: DEPRECATED. Whether to clear KV and prefix caches clear_cache: DEPRECATED. Whether to clear KV and prefix caches
after draining. after draining.
@@ -4,7 +4,6 @@ from collections.abc import Sequence
from typing import Any from typing import Any
from vllm import PoolingParams, PoolingRequestOutput from vllm import PoolingParams, PoolingRequestOutput
from vllm.logger import init_logger
from vllm.plugins.io_processors import get_io_processor from vllm.plugins.io_processors import get_io_processor
from vllm.renderers.inputs.preprocess import parse_model_prompt, prompt_to_seq from vllm.renderers.inputs.preprocess import parse_model_prompt, prompt_to_seq
@@ -21,8 +20,6 @@ from ..typing import (
) )
from .protocol import IOProcessorRequest, IOProcessorResponse from .protocol import IOProcessorRequest, IOProcessorResponse
logger = init_logger(__name__)
class UnsupportedCombinedTaskIOProcessor(PoolingIOProcessor): class UnsupportedCombinedTaskIOProcessor(PoolingIOProcessor):
name = "embed&token_classify" name = "embed&token_classify"
@@ -117,23 +114,7 @@ class PluginWithIOProcessorPlugins(PoolingIOProcessor):
ctx.final_res_batch, ctx.final_res_batch,
request_id=ctx.request_id, request_id=ctx.request_id,
) )
ctx.response = IOProcessorResponse(request_id=ctx.request_id, data=output)
if callable(
output_to_response := getattr(self.io_processor, "output_to_response", None)
):
logger.warning_once(
"`IOProcessor.output_to_response` is deprecated. To ensure "
"consistency between offline and online APIs, "
"`IOProcessorResponse` will become a transparent wrapper "
"around output data from v0.19 onwards.",
)
if hasattr(output, "request_id") and output.request_id is None:
output.request_id = ctx.request_id # type: ignore
ctx.response = output_to_response(output) # type: ignore
else:
ctx.response = IOProcessorResponse(request_id=ctx.request_id, data=output)
####################################### #######################################
# offline APIs # offline APIs
-3
View File
@@ -32,7 +32,6 @@ router = APIRouter()
async def pause_generation( async def pause_generation(
raw_request: Request, raw_request: Request,
mode: Annotated[PauseMode, Query()] = "abort", mode: Annotated[PauseMode, Query()] = "abort",
wait_for_inflight_requests: bool = Query(False),
clear_cache: Annotated[bool, Query()] = True, clear_cache: Annotated[bool, Query()] = True,
) -> JSONResponse: ) -> JSONResponse:
"""Pause generation requests to allow weight updates. """Pause generation requests to allow weight updates.
@@ -44,7 +43,6 @@ async def pause_generation(
- ``"abort"``: Abort all in-flight requests immediately (default). - ``"abort"``: Abort all in-flight requests immediately (default).
- ``"wait"``: Wait for in-flight requests to complete. - ``"wait"``: Wait for in-flight requests to complete.
- ``"keep"``: Freeze requests in queue; they resume on /resume. - ``"keep"``: Freeze requests in queue; they resume on /resume.
wait_for_inflight_requests: DEPRECATED. Use ``mode="wait"`` instead.
clear_cache: DEPRECATED. Whether to clear KV/prefix caches after clear_cache: DEPRECATED. Whether to clear KV/prefix caches after
draining. Ignored when mode="keep". draining. Ignored when mode="keep".
@@ -55,7 +53,6 @@ async def pause_generation(
await engine.pause_generation( await engine.pause_generation(
mode=mode, mode=mode,
clear_cache=clear_cache, clear_cache=clear_cache,
wait_for_inflight_requests=wait_for_inflight_requests,
) )
return JSONResponse( return JSONResponse(
content={"status": "paused"}, content={"status": "paused"},
-17
View File
@@ -1,17 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from vllm.entrypoints.cli.main import main as vllm_main
from vllm.logger import init_logger
logger = init_logger(__name__)
# Backwards compatibility for the move from vllm.scripts to
# vllm.entrypoints.cli.main
def main():
logger.warning(
"vllm.scripts.main() is deprecated. Please re-install "
"vllm or use vllm.entrypoints.cli.main.main() instead."
)
vllm_main()
-12
View File
@@ -4,7 +4,6 @@ import asyncio
import os import os
import socket import socket
import time import time
import warnings
from collections.abc import AsyncGenerator, Iterable, Mapping from collections.abc import AsyncGenerator, Iterable, Mapping
from copy import copy from copy import copy
from typing import Any from typing import Any
@@ -944,7 +943,6 @@ class AsyncLLM(EngineClient):
self, self,
*, *,
mode: PauseMode = "abort", mode: PauseMode = "abort",
wait_for_inflight_requests: bool | None = None,
clear_cache: bool = True, clear_cache: bool = True,
) -> None: ) -> None:
"""Pause generation to allow model weight updates. """Pause generation to allow model weight updates.
@@ -960,20 +958,10 @@ class AsyncLLM(EngineClient):
- ``"wait"``: Wait for in-flight requests to complete. - ``"wait"``: Wait for in-flight requests to complete.
- ``"keep"``: Freeze requests in queue; they resume on - ``"keep"``: Freeze requests in queue; they resume on
:meth:`resume_generation`. :meth:`resume_generation`.
wait_for_inflight_requests: DEPRECATED: use mode argument.
clear_cache: Whether to clear KV cache and prefix cache after clear_cache: Whether to clear KV cache and prefix cache after
draining. Set to ``False`` to preserve cache for faster resume. draining. Set to ``False`` to preserve cache for faster resume.
""" """
if wait_for_inflight_requests:
warnings.warn(
"The `wait_for_inflight_requests` parameter in "
"`AsyncLLM.pause_generation()` is deprecated. "
"Please use `mode` argument instead.",
DeprecationWarning,
stacklevel=2,
)
mode = "wait"
if clear_cache: if clear_cache:
await self.renderer.clear_mm_cache_async() await self.renderer.clear_mm_cache_async()
await self.engine_core.pause_scheduler_async(mode=mode, clear_cache=clear_cache) await self.engine_core.pause_scheduler_async(mode=mode, clear_cache=clear_cache)