48 lines
1.7 KiB
Python
48 lines
1.7 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
from __future__ import annotations
|
|
|
|
from vllm.logger import init_logger
|
|
from vllm.v1.core.sched.output import SchedulerOutput
|
|
from vllm.v1.core.sched.scheduler import Scheduler
|
|
from vllm.v1.request import Request, RequestStatus
|
|
|
|
logger = init_logger(__name__)
|
|
|
|
|
|
class AsyncScheduler(Scheduler):
|
|
|
|
def _update_after_schedule(
|
|
self,
|
|
scheduler_output: SchedulerOutput,
|
|
) -> None:
|
|
super()._update_after_schedule(scheduler_output)
|
|
for req_id in scheduler_output.num_scheduled_tokens:
|
|
request = self.requests[req_id]
|
|
if (request.num_computed_tokens == request.num_tokens +
|
|
request.num_output_placeholders):
|
|
# The request will generate a new token in this scheduling step.
|
|
# TODO(woosuk): Support speculative decoding.
|
|
request.num_output_placeholders += 1
|
|
|
|
def _update_request_with_output(
|
|
self,
|
|
request: Request,
|
|
new_token_ids: list[int],
|
|
) -> tuple[list[int], bool]:
|
|
status_before_update = request.status
|
|
new_token_ids, stopped = super()._update_request_with_output(
|
|
request, new_token_ids)
|
|
|
|
# Update the number of output placeholders.
|
|
request.num_output_placeholders -= len(new_token_ids)
|
|
assert request.num_output_placeholders >= 0
|
|
|
|
# Cache the new tokens. Preempted requests should be skipped.
|
|
if status_before_update == RequestStatus.RUNNING:
|
|
self.kv_cache_manager.cache_blocks(
|
|
request,
|
|
request.num_computed_tokens - request.num_output_placeholders)
|
|
return new_token_ids, stopped
|