System Prompt and User Instruction

app.py

@@ -85,11 +85,21 @@ model = Mistral3ForConditionalGeneration.from_pretrained(
     ).eval()
 
 
-SYSTEM_PROMPT_TEXT = (
-    "You are a world-class geolocation expert. Given a street-view style image, "
-    "think step by step about visual clues and infer approximate coordinates. "
-    "When you conclude, output your answer inside [ANSWER]lat,lng[/ANSWER]."
-)
+# SYSTEM_PROMPT_TEXT = (
+#     "You are a world-class geolocation expert. Given a street-view style image, "
+#     "think step by step about visual clues and infer approximate coordinates. "
+#     "When you conclude, output your answer inside [ANSWER]lat,lng[/ANSWER]."
+# )
+SYSTEM_PROMPT_TEXT = """# HOW YOU SHOULD THINK AND ANSWER
+First draft your thinking process (inner monologue) until you arrive at a response. Format your response using Markdown, and use LaTeX for any mathematical equations. Write both your thoughts and the response in the same language as the input.
+
+Your thinking process must follow the template below:[THINK]Your thoughts or/and draft, like working through an exercise on scratch paper. Be as casual and as long as you want until you are confident to generate the response to the user.[/THINK]Here, provide a self-contained response."""
+
+USER_INSTRUCTION = """You are a world-class geolocation expert. Given a street-view style image, think step by step about visual clues and infer approximate coordinates.
+When you conclude, output your final answer inside [ANSWER]lat,lng[/ANSWER].
+
+Please analyze this image and provide coordinates in the required format."""
+
 @spaces.GPU(duration=120)
 def llm_decode_image_return_text(image_bytes: bytes) -> str:
     print(f"[llm] decode start. image_bytes={len(image_bytes)} bytes")
@@ -101,7 +111,7 @@ def llm_decode_image_return_text(image_bytes: bytes) -> str:
         messages = [
             {"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT_TEXT}]},
             {"role": "user", "content": [
-                {"type": "text", "text": "Please analyze this image and provide coordinates."},
+                {"type": "text", "text": USER_INSTRUCTION},
                 {"type": "image_url", "image_url": {"url": data_url}},
             ]},
         ]
@@ -140,7 +150,7 @@ def llm_stream_image_text(image_bytes: bytes):
         messages = [
             {"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT_TEXT}]},
             {"role": "user", "content": [
-                {"type": "text", "text": "Please analyze this image and provide coordinates."},
+                {"type": "text", "text": USER_INSTRUCTION},
                 {"type": "image_url", "image_url": {"url": data_url}},
             ]},
         ]