fix: increase ontology generation output budget and harden JSON parsing

backend/app/services/ontology_generator.py

@@ -197,7 +197,7 @@ class OntologyGenerator:
         result = self.llm_client.chat_json(
             messages=messages,
             temperature=0.3,
-            max_tokens=4096
+            max_tokens=32768
         )
         
         # 验证和后处理

backend/app/utils/llm_client.py

@@ -54,9 +54,9 @@ class LLMClient:
         kwargs = {
             "model": self.model,
             "messages": messages,
-            "temperature": temperature,
-            "max_tokens": max_tokens,
+            "max_completion_tokens": max_tokens,
         }
+        kwargs["temperature"] = temperature
         
         if response_format:
             kwargs["response_format"] = response_format
@@ -71,7 +71,7 @@ class LLMClient:
         self,
         messages: List[Dict[str, str]],
         temperature: float = 0.3,
-        max_tokens: int = 4096
+        max_tokens: int = 16384
     ) -> Dict[str, Any]:
         """
         发送聊天请求并返回JSON
@@ -88,7 +88,6 @@ class LLMClient:
             messages=messages,
             temperature=temperature,
             max_tokens=max_tokens,
-            response_format={"type": "json_object"}
         )
         # 清理markdown代码块标记
         cleaned_response = response.strip()
@@ -99,5 +98,27 @@ class LLMClient:
         try:
             return json.loads(cleaned_response)
         except json.JSONDecodeError:
-            raise ValueError(f"LLM返回的JSON格式无效: {cleaned_response}")
+            # Aggressive JSON repair for truncated responses
+            repaired = cleaned_response
+            # Try progressively trimming from the end until valid JSON
+            for trim in range(min(len(repaired), 2000)):
+                candidate = repaired[:len(repaired) - trim].rstrip(', \n\t\r')
+                # Remove incomplete string at end
+                if candidate and candidate[-1] not in ']}}"':
+                    continue
+                open_braces = candidate.count('{') - candidate.count('}')
+                open_brackets = candidate.count('[') - candidate.count(']')
+                fixed = candidate
+                for _ in range(open_brackets):
+                    fixed += ']'
+                for _ in range(open_braces):
+                    fixed += '}'
+                try:
+                    result = json.loads(fixed)
+                    import logging
+                    logging.getLogger(__name__).warning(f"JSON repaired by trimming {trim} chars")
+                    return result
+                except json.JSONDecodeError:
+                    continue
+            raise ValueError(f"LLM返回的JSON格式无效: {cleaned_response[:500]}")
 

frontend/package-lock.json

@@ -1331,7 +1331,6 @@
       "resolved": "https://registry.npmjs.org/d3-selection/-/d3-selection-3.0.0.tgz",
       "integrity": "sha512-fmTRWbNMmsmWq6xJV8D19U/gw/bwrHfNXxrIN+HfZgnzqTHp9jOmKMhsTUjXOJnZOdZY9Q28y4yebKzqDKlxlQ==",
       "license": "ISC",
-      "peer": true,
       "engines": {
         "node": ">=12"
       }
@@ -1809,7 +1808,6 @@
       "integrity": "sha512-5gTmgEY/sqK6gFXLIsQNH19lWb4ebPDLA4SdLP7dsWkIXHWlG66oPuVvXSGFPppYZz8ZDZq0dYYrbHfBCVUb1Q==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "engines": {
         "node": ">=12"
       },
@@ -1943,7 +1941,6 @@
       "integrity": "sha512-ITcnkFeR3+fI8P1wMgItjGrR10170d8auB4EpMLPqmx6uxElH3a/hHGQabSHKdqd4FXWO1nFIp9rRn7JQ34ACQ==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "esbuild": "^0.25.0",
         "fdir": "^6.5.0",
@@ -2018,7 +2015,6 @@
       "resolved": "https://registry.npmjs.org/vue/-/vue-3.5.25.tgz",
       "integrity": "sha512-YLVdgv2K13WJ6n+kD5owehKtEXwdwXuj2TTyJMsO7pSeKw2bfRNZGjhB7YzrpbMYj5b5QsUebHpOqR3R3ziy/g==",
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "@vue/compiler-dom": "3.5.25",
         "@vue/compiler-sfc": "3.5.25",