xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-21 22:44:31 +08:00
Code Issues Packages Projects Releases Wiki Activity

remove the constraint for SD

Browse Source 
Signed-off-by: Yu Gong <yu3.gong@gmail.com>
This commit is contained in:
Yu Gong 2025-12-23 20:00:19 +00:00
parent e95e2145c5
commit 094eaef7b3
1 changed files with 0 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/v1/cudagraph_dispatcher.py
View File

@ -194,15 +194,7 @@ class CudagraphDispatcher:
): ):
return CUDAGraphMode.NONE, BatchDescriptor(num_tokens) return CUDAGraphMode.NONE, BatchDescriptor(num_tokens)
# When speculative decoding is enabled, always use max_loras for lookup
# since we only capture graphs with max_loras
effective_num_active_loras = num_active_loras effective_num_active_loras = num_active_loras
if (
self.vllm_config.speculative_config is not None
and self.vllm_config.lora_config is not None
and has_lora
):
effective_num_active_loras = self.vllm_config.lora_config.max_loras
batch_desc = self._create_padded_batch_descriptor( batch_desc = self._create_padded_batch_descriptor(
num_tokens, uniform_decode, has_lora, effective_num_active_loras num_tokens, uniform_decode, has_lora, effective_num_active_loras