biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Config] add "qwen" as a native eagle3 target supported model (#22333)

Browse Source 
Signed-off-by: lechen <lecself@163.com>
Signed-off-by: LeChen <lecself@163.com>
This commit is contained in:
Le Chen
2025-08-10 11:21:05 +08:00
committed by GitHub
parent 0c5254b82a
commit 3d7363e61c
5 changed files with 30 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/config/__init__.py
View File

@@ -2852,13 +2852,7 @@ class SpeculativeConfig:
"speculative decoding is > 1, but got "
f"{self.disable_by_batch_size=}")
from vllm.transformers_utils.configs import SpeculatorsConfig
eagle3_target_supported = ["llama"]
if self.draft_model_config and isinstance(
self.draft_model_config.hf_config, SpeculatorsConfig):
eagle3_target_supported.append("qwen")
eagle3_target_supported = ["llama", "qwen"]
if self.method == "eagle3" and self.target_model_config and not any(
supported_model in
self.target_model_config.hf_text_config.model_type