Skip to content

Commit f067623

Browse files
GDzhu01MengqingCaowangxiyuan
authored
[Bugfix] fix mtp and eagle aclgraph bug (#4710)
### What this PR does / why we need it? fix mtp and eagle aclgraph bug - vLLM version: v0.12.0 - vLLM main: vllm-project/vllm@ad32e3e Signed-off-by: GDzhu01 <[email protected]> Co-authored-by: Mengqing Cao <[email protected]> Co-authored-by: wangxiyuan <[email protected]>
1 parent 7403399 commit f067623

File tree

2 files changed

+2
-2
lines changed

2 files changed

+2
-2
lines changed

vllm_ascend/spec_decode/eagle_proposer.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -51,7 +51,7 @@ def __init__(self,
5151
not self.vllm_config.model_config.enforce_eager)
5252

5353
self.cudagraph_batch_sizes = list(
54-
reversed(
54+
sorted(
5555
self.vllm_config.compilation_config.cudagraph_capture_sizes))
5656

5757
# persistent buffers for cuda graph

vllm_ascend/spec_decode/mtp_proposer.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -108,7 +108,7 @@ def __init__(
108108
self.use_aclgraph = self.runner._use_aclgraph()
109109

110110
self.cudagraph_batch_sizes = (list(
111-
reversed(
111+
sorted(
112112
self.vllm_config.compilation_config.cudagraph_capture_sizes))
113113
if self.use_aclgraph else [])
114114

0 commit comments

Comments
 (0)