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

drop ascend scheduler (#4498)

Browse Source 
Ascend scheduler was added for non chunk prefill case before, since that
the npu ops didn't work well with chunked prefill.

Now the ops with chunked prefill work better, it's time to remove the
ascend scheduler to use vLLM default scheduler.

- vLLM version: v0.11.2

---------

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-11-29 16:18:34 +08:00
committed by GitHub
parent 53a52d6614
commit f10acddb78
52 changed files with 85 additions and 2948 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
vllm_ascend/ascend_config.py
View File

@@ -39,11 +39,6 @@ class AscendConfig:
self.torchair_graph_config = TorchairGraphConfig(
torchair_graph_config, vllm_config, additional_config)
ascend_scheduler_config = additional_config.get(
"ascend_scheduler_config", {})
self.ascend_scheduler_config = AscendSchedulerConfig(
ascend_scheduler_config)
# Dump / PrecisionDebugger configuration
dump_config_path = additional_config.get("dump_config", None)
self.dump_config = DumpConfig(dump_config_path)
@@ -220,20 +215,6 @@ class TorchairGraphConfig:
)
class AscendSchedulerConfig:
"""
Configuration Object for ascend_scheduler_config from additional_config
"""
def __init__(self, ascend_scheduler_config: dict):
self.enabled = ascend_scheduler_config.get("enabled", False)
# Ascend scheduler is based on vllm v0 scheduler, so we should support
# all vllm v0 scheduler configs as well.
for k, v in ascend_scheduler_config.items():
if not hasattr(self, k):
setattr(self, k, v)
class DumpConfig:
"""
Configuration object for dump/PrecisionDebugger settings.