feat(config): update default LLM model configuration

- Changed default model from qwen3-30b-a3b-instruct-2507 to qwen-flash
- Updated evaluation tools to use EVALUATION_PROMPT_FOR_QUESTION instead of QUESTION2
- Added new PROMPT_MEMZERO_JSON2 configuration with context priority rules
- Modified llm_request_for_json to use qwen-flash as default model
- Updated halumem evaluation to specify qwen3-max model explicitly for certain requests

bench/halumem/eval_tools.py

@@ -120,7 +120,8 @@ async def evaluation_for_question2(
     dialogue: The formatted dialogue history (role, content, time_created).
     """
 
-    prompt = _PROMPTS["EVALUATION_PROMPT_FOR_QUESTION2"].format(
+    # prompt = _PROMPTS["EVALUATION_PROMPT_FOR_QUESTION2"].format(
+    prompt = _PROMPTS["EVALUATION_PROMPT_FOR_QUESTION"].format(
         question=question,
         reference_answer=reference_answer,
         key_memory_points=key_memory_points,
@@ -128,7 +129,7 @@ async def evaluation_for_question2(
         dialogue=dialogue,
     )
 
-    result = await llm_request_for_json(prompt)
+    result = await llm_request_for_json(prompt, model_name="qwen3-max")
 
     return result
 

bench/halumem/halumem.yaml

@@ -26,6 +26,30 @@ PROMPT_MEMZERO_JSON: |
   }}
   ```
 
+PROMPT_MEMZERO_JSON2: |
+  # CONTEXT:
+  {context}
+
+  # CONTEXT PRIORITY:
+  When the context contains information from multiple sources, follow this strict priority order:
+  1. **Historical Dialogue** (highest priority) - Direct conversation content
+  2. **Extracted Memories** (medium priority) - Summarized memory points
+  3. **User Profile** (lowest priority) - General user information
+  
+  # Question:
+  {question}
+
+  # OUTPUT FORMAT:
+  Do not hallucinate; strictly answer the user's question based on the content of the CONTEXT.
+  Please provide your response in the following JSON format:
+
+  ```json
+  {{
+    "reasoning": "reasoning content",
+    "answer": "Provide a detailed answer"
+  }}
+  ```
+
 PROMPT_MEMZERO: |
   You are an intelligent memory assistant tasked with retrieving accurate information from conversation memories.
 

bench/halumem/llms.py

@@ -58,7 +58,8 @@ async def llm_request(prompt, model_name: str = "qwen3-max", **kwargs) -> str:
     reraise=True,
     before_sleep=before_sleep_log(logger, logging.WARNING),
 )
-async def llm_request_for_json(prompt, model_name: str = "qwen3-max", **kwargs):
+async def llm_request_for_json(prompt, model_name: str = "qwen-flash", **kwargs):
+    # async def llm_request_for_json(prompt, model_name: str = "qwen3-max", **kwargs):
     """Make an LLM request expecting JSON response using ReMe's LLM.
     
     Args:

reme_ai/core/config/default.yaml

@@ -15,7 +15,8 @@ http:
 llm:
   default:
     backend: openai
-    model_name: qwen3-30b-a3b-instruct-2507
+#    model_name: qwen3-30b-a3b-instruct-2507
+    model_name: qwen-flash
     request_interval: 1
     temperature: 0.0001