EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Drop ascend scheduler (#4623)

Browse Source 
It's safe to drop ascend scheduler now. The related test and doc has
been removed already


- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-12-05 09:03:45 +08:00
committed by GitHub
parent 00b4fb80de
commit ea54388e19
12 changed files with 34 additions and 767 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
vllm_ascend/platform.py
View File

@@ -30,6 +30,7 @@ from vllm_ascend.ascend_config import (check_ascend_config, get_ascend_config,
init_ascend_config)
from vllm_ascend.torchair.utils import (check_torchair_cache_exist,
delete_torchair_cache_file)
from vllm_ascend.utils import refresh_block_size
# isort: off
from vllm_ascend.utils import (
@@ -160,7 +161,6 @@ class NPUPlatform(Platform):
model_config = vllm_config.model_config
parallel_config = vllm_config.parallel_config
cache_config = vllm_config.cache_config
ascend_scheduler_config = ascend_config.ascend_scheduler_config
ascend_compilation_config = ascend_config.ascend_compilation_config
if ascend_compilation_config:
vllm_config.additional_config.setdefault(
@@ -307,38 +307,13 @@ class NPUPlatform(Platform):
else:
parallel_config.worker_cls = "vllm_ascend.worker.worker_v1.NPUWorker"
if cache_config:
if cache_config.block_size is None:
cache_config.block_size = 128
if cache_config.enable_prefix_caching or \
not ascend_scheduler_config.enabled or \
getattr(ascend_scheduler_config, "enable_chunked_prefill", False):
logger.warning(
"If chunked prefill or prefix caching is enabled, block size must be set to 128."
)
origin_block_size = cache_config.block_size
cache_config.block_size = 128
# TODO(MengqingCao): Remove the model_type check, after resolving the hidden error in get_kv_cache_groups.
if model_config and model_config.hf_config.model_type == "qwen3_next":
logger.warning(
"When running qwen3-next model, block_size needs to be restored to its original value."
)
cache_config.block_size = origin_block_size
refresh_block_size(vllm_config)
# Activate custom ops for v1, except on 310P
if get_ascend_device_type() != AscendDeviceType._310P:
compilation_config.custom_ops = ["all"]
# If ascend_scheduler_config is enabled,
# extents original scheduler_config to use AscendScheduler.
if ascend_config.ascend_scheduler_config.enabled:
from vllm_ascend.core.schedule_config import AscendSchedulerConfig
ascend_scheduler_config = AscendSchedulerConfig.initialize_from_config(
vllm_config.scheduler_config,
ascend_config.ascend_scheduler_config)
vllm_config.scheduler_config = ascend_scheduler_config
elif ascend_config.recompute_scheduler_enable:
if ascend_config.recompute_scheduler_enable:
from vllm_ascend.core.recompute_schedule_config import \
RecomputeSchedulerConfig
recompute_scheduler_config = RecomputeSchedulerConfig.initialize_from_config(