xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-21 17:55:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[misc] refactor speculative config (#25657)

Browse Source 
Signed-off-by: zxw <1020938856@qq.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
yyzxw 2025-09-26 16:22:06 +08:00 committed by yewentao256
parent bc37468b3c
commit c761b84d5f
1 changed files with 6 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/config/speculative.py
View File

@ -209,12 +209,9 @@ class SpeculativeConfig:
if self.model is None and self.num_speculative_tokens is not None:
# TODO(Shangming): Refactor mtp configuration logic when supporting
# mtp acceleration for more models besides deepseek_v3
if self.target_model_config and \
(self.target_model_config.hf_text_config.model_type \
== "deepseek_v3" or
self.target_model_config.hf_text_config.model_type in
("mimo","ernie4_5_moe", "qwen3_next")):
if (self.target_model_config
and self.target_model_config.hf_text_config.model_type
in ("deepseek_v3", "mimo", "ernie4_5_moe", "qwen3_next")):
# use the draft model from the same model:
self.model = self.target_model_config.model
# Align the quantization of draft model for cases such as
@ -224,8 +221,9 @@ class SpeculativeConfig:
elif self.method in ("ngram", "[ngram]"):
self.model = "ngram"
else:
raise ValueError("num_speculative_tokens was provided without "
"speculative model.")
raise ValueError(
"num_speculative_tokens was provided but without "
"speculative model.")
# Automatically configure the method for ngram when "model" is used
# instead of "method"