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

7
vllm_ascend/worker/model_runner_v1.py
View File

@@ -332,10 +332,6 @@ class NPUModelRunner(LoRAModelRunnerMixin, ECConnectorModelRunnerMixin):
# Ascend-specific configurations
self.ascend_config = get_ascend_config()
if self.ascend_config.ascend_scheduler_config.enabled:
self.chunked_prefill_enabled = self.scheduler_config.enable_chunked_prefill
else:
self.chunked_prefill_enabled = True
self.weight_prefetch_method = WeightPrefetchMethod(
self.ascend_config.weight_prefetch_config)
# Dump / PrecisionDebugger configuration now comes from AscendConfig
@@ -1932,7 +1928,6 @@ class NPUModelRunner(LoRAModelRunnerMixin, ECConnectorModelRunnerMixin):
def _build_attn_state(self, num_reqs, num_scheduled_tokens,
num_valid_tokens):
ascend_config = get_ascend_config()
if np.array_equal(self.seq_lens_np[:num_reqs], num_scheduled_tokens):
attn_state = AscendAttentionState.PrefillNoCache
# We assume it is the decode stage, where prefill occurs but only one token is not hit in cache.
@@ -1949,7 +1944,7 @@ class NPUModelRunner(LoRAModelRunnerMixin, ECConnectorModelRunnerMixin):
else:
attn_state = AscendAttentionState.ChunkedPrefill
# splitfuse
elif not ascend_config.ascend_scheduler_config.enabled or self.chunked_prefill_enabled:
elif self.scheduler_config.enable_chunked_prefill:
attn_state = AscendAttentionState.ChunkedPrefill
else:
attn_state = AscendAttentionState.PrefillCacheHit