Optimize retract (#440)

python/sglang/global_config.py

@@ -28,5 +28,11 @@ class GlobalConfig:
         # Request dependency time due to network delay
         self.request_dependency_time = 0.03
 
+        # New generation token ratio estimation
+        self.base_new_token_ratio = 0.4
+        self.base_min_new_token_ratio = 0.2
+        self.new_token_ratio_decay = 0.0001
+        self.new_token_ratio_recovery = 0.05
+
 
 global_config = GlobalConfig()