improve processing

app.py

@@ -59,7 +59,6 @@ class BasicAgent:
         llm = HuggingFaceInferenceAPI(
             model_name="Qwen/Qwen2.5-Coder-32B-Instruct",
             token = hf_api_key,
-            max_tokens=32767
         )
 
         # Create agent with function-calling-compatible LLM
@@ -77,6 +76,12 @@ class BasicAgent:
         )
         print("✅ BasicAgent initialized.")
 
+    async def answer_once(self, prompt: str) -> str:
+        """Ask one question, then clear the dialogue so tokens never pile up."""
+        reply = await self.agent.run(prompt)
+        self.agent.reset()          # <<< flush conversation context
+        return reply
+
     async def __call__(self, input_text: str):
         # Use run() within async context
         return await self.agent.run(input_text)
@@ -149,7 +154,7 @@ async def run_and_submit_all(profile: gr.OAuthProfile | None):
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         try:
-            submitted_answer = await agent(question_text)
+            submitted_answer = await agent.answer_once(question_text)
             answers_payload.append(
                 {"task_id": task_id, "submitted_answer": submitted_answer}
             )