Support limit thinking lengths (#4244)

Co-authored-by: K11OntheBoat <“ruianmaidanglao@163.com”>
2025-10-06 17:17:14 +08:00 · 2025-09-24 17:30:53 +08:00
parent 12043fc476
commit 05b7800d80
8 changed files with 184 additions and 26 deletions
--- a/fastdeploy/config.py
+++ b/fastdeploy/config.py
@@ -190,6 +190,7 @@ class ModelConfig:
        self.reasoning_parser = None
        self.pad_token_id: int = -1
        self.eos_tokens_lens: int = 2
+        self.think_end_id = None
        self.lm_head_fp32: bool = False
        self.model_format = "auto"
        self.runner = "auto"