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

[bugfix] fix typo of _skip_all_reduce_across_dp_group (#5435)

Browse Source 
### What this PR does / why we need it?
 fix typo of _skip_all_reduce_across_dp_group
### Does this PR introduce _any_ user-facing change?
no

### How was this patch tested?
- vLLM version: release/v0.13.0
- vLLM main:
81786c8774

Signed-off-by: jiangkuaixue123 <jiangxiaozhou111@163.com>
This commit is contained in:
jiangkuaixue123
2025-12-27 17:50:04 +08:00
committed by GitHub
parent c30c3dc831
commit e91e11d3b0
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm_ascend/worker/model_runner_v1.py
View File

@@ -397,7 +397,7 @@ class NPUModelRunner(GPUModelRunner):
def _use_aclgraph(self) -> bool: def _use_aclgraph(self) -> bool:
return self.compilation_config.cudagraph_mode != CUDAGraphMode.NONE and self.compilation_config.mode == CompilationMode.VLLM_COMPILE and not self.model_config.enforce_eager return self.compilation_config.cudagraph_mode != CUDAGraphMode.NONE and self.compilation_config.mode == CompilationMode.VLLM_COMPILE and not self.model_config.enforce_eager
def _skip_all_reduce_acorss_dp_group(self) -> bool: def _skip_all_reduce_across_dp_group(self) -> bool:
""" """
Decide whether to skip the all-reduce across the data-parallel (DP) group. Decide whether to skip the all-reduce across the data-parallel (DP) group.
@@ -446,7 +446,7 @@ class NPUModelRunner(GPUModelRunner):
if self.dp_size == 1: if self.dp_size == 1:
return num_tokens, None, with_prefill return num_tokens, None, with_prefill
if self._skip_all_reduce_acorss_dp_group(): if self._skip_all_reduce_across_dp_group():
num_tokens_after_padding = torch.tensor([num_tokens] * num_tokens_after_padding = torch.tensor([num_tokens] *
self.dp_size, self.dp_size,
device="cpu", device="cpu",