Disable thinking mode for faster chat responses

3 weeks ago · fe6f429e21
parent 7731595022
commit fe6f429e21
1 changed files with 1 additions and 0 deletions
--- a/services/rag-api/services/chat.py
+++ b/services/rag-api/services/chat.py
@ -54,6 +54,7 @@ Answer based on the above context:"""
    payload = {
        'model': model,
        'stream': True,
        'think': False,  # Disable thinking mode for faster responses
        'messages': [
            {'role': 'system', 'content': SYSTEM_PROMPT},
            {'role': 'user', 'content': prompt},