Update inference.py

fix the thinking budgets

Author: codelion

optillm/inference.py

@@ -1613,31 +1613,31 @@ def get_effort_profile(reasoning_effort: str, max_tokens: int = 4096) -> dict:
                 "However,",
                 "Additionally,",
             ],
-            "prefill": "Let me think about this briefly..."
+            "prefill": ""
         },
         "medium": {
             "min_tokens_pct": 0.10,  
             "max_tokens_pct": 0.66,  # 66% of max_tokens
-            "max_thoughts": 128,
+            "max_thoughts": 256,
             "thought_switch_tokens": [
                 "Wait,",
                 "Alternatively,",
                 "However,",
                 "Additionally,",
             ],
-            "prefill": "Let me analyze this from multiple angles..."
+            "prefill": ""
         },
         "high": {
             "min_tokens_pct": 0.10,  
             "max_tokens_pct": 0.90,  # 90% of max_tokens
-            "max_thoughts": 256,
+            "max_thoughts": 512,
             "thought_switch_tokens": [
                 "Wait,",
                 "Alternatively,",
                 "However,",
                 "Additionally,",
             ],
-            "prefill": "This requires careful analysis. Let me think through it systematically..."
+            "prefill": ""
         }
     }