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

[MISC] Clean up torch_npu (#688)

Browse Source 
torch_npu 2.5.1 support autoload now. This patch does:
1. remove useless torch_npu import
2. replace `torch_npu.npu` to `torch.npu`.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-04-29 18:03:38 +08:00
committed by GitHub
parent 0329fad927
commit b917361ca5
15 changed files with 18 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm_ascend/worker/pooling_model_runner.py
View File

@@ -134,9 +134,8 @@ class NPUPoolingModelRunner(
} if self.has_inner_state else {}
if (self.observability_config is not None
and self.observability_config.collect_model_forward_time):
import torch_npu
model_forward_start = torch_npu.npu.Event(enable_timing=True)
model_forward_end = torch_npu.npu.Event(enable_timing=True)
model_forward_start = torch.npu.Event(enable_timing=True)
model_forward_end = torch.npu.Event(enable_timing=True)
model_forward_start.record()
cross_enc_kwargs = {}