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

[BugFix] Fix some issues caused by the ascending order of cudagraph_capture_sizes (#4338)

Browse Source 
### What this PR does / why we need it?
In [#26016](https://github.com/vllm-project/vllm/pull/26016), vllm
change the `cudagraph_capture_sizes` to be in ascending order. This PR
fixes related issues caused by this.
### Does this PR introduce _any_ user-facing change?
No.

### How was this patch tested?


- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

---------

Signed-off-by: Angazenn <supperccell@163.com>
This commit is contained in:
Angazenn
2025-11-22 17:33:12 +08:00
committed by GitHub
parent fff258bce1
commit 9b3a484b46
3 changed files with 13 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm_ascend/worker/model_runner_v1.py
View File

@@ -664,7 +664,11 @@ class NPUModelRunner(LoRAModelRunnerMixin):
# tokens is less than or equal to mc2_tokens_capacity. According to _set_cudagraph_sizes,
# the max number of tokens in graph is min(max_num_seqs * uniform_decode_query_len, 512).
if self.compilation_config.cudagraph_capture_sizes:
max_num_tokens = self.compilation_config.cudagraph_capture_sizes[0]
if vllm_version_is("0.11.0"):
max_num_tokens = self.compilation_config.cudagraph_capture_sizes[
0]
else:
max_num_tokens = self.compilation_config.max_cudagraph_capture_size
else:
# NOTE: To save memory, we cap the max number of tokens to 512.
max_num_tokens = min(