feat: update model name

app/services/ocr_service.py

@@ -527,7 +527,7 @@ class MineruOCRService(OCRServiceBase):
             messages = [{"role": "user", "content": [{"type": "image_url", "image_url": {"url": image_url}}, {"type": "text", "text": prompt}]}]
 
             response = self.openai_client.chat.completions.create(
-                model="PaddlePaddle/PaddleOCR-VL",
+                model="PaddleOCR-VL-0.9B",  # Use exact model name from vLLM server
                 messages=messages,
                 temperature=0.0,
             )
@@ -553,31 +553,15 @@ class MineruOCRService(OCRServiceBase):
         if not image_pattern.search(markdown_content):
             return markdown_content
 
-        print(f"[DEBUG] Found image reference in markdown, triggering PaddleOCR-VL recognition")
+        formula_text = self._recognize_formula_with_paddleocr_vl(original_image)
 
-        try:
-            # For now, use the entire image for formula recognition
-            # TODO: Extract specific regions if image paths contain coordinates
-            formula_text = self._recognize_formula_with_paddleocr_vl(original_image)
+        if formula_text.startswith("\[") or formula_text.startswith("\("):
+            formula_text = formula_text.replace("\[", "$$").replace("\(", "$$")
+            formula_text = formula_text.replace("\]", "$$").replace("\)", "$$")
+        else:
+            formula_text = f"$${formula_text}$$"
 
-            print(f"[DEBUG] PaddleOCR-VL recognized formula: {formula_text[:100] if formula_text else 'Empty'}...")
-
-            # Replace image references with recognized formulas
-            # Wrap in display math delimiters if not already wrapped
-            if formula_text and not formula_text.startswith("$$"):
-                formula_text = f"$${formula_text}$$"
-
-            markdown_content = image_pattern.sub(formula_text, markdown_content)
-            print(f"[DEBUG] Formula recognition successful, updated markdown")
-
-        except Exception as e:
-            # If formula recognition fails, keep original content and log error
-            import traceback
-
-            print(f"[ERROR] Formula recognition failed: {e}")
-            print(f"[ERROR] Traceback: {traceback.format_exc()}")
-
-        return markdown_content
+        return formula_text
 
     def recognize(self, image: np.ndarray) -> dict:
         """Recognize content using local file_parse API.