Add error handling for LLM requests

main.py

@@ -322,9 +322,12 @@ async def __query_ml_predict(qtype: QType, content: str, header: str, token_limi
                     timeout=None
             )
 
-            _w_content = _w.json()['choices'][0]['message']['content'].split('\n\n',1)[1]
-            print(time() - stopwatch)
-            return {"content": f"{_r_content}\n{_w_content}", "style": QType.STMT}
+            try:
+                _w_content = _w.json()['choices'][0]['message']['content'].split('\n\n',1)[1]
+                #print(time() - stopwatch)
+                return {"content": f"{_r_content}\n{_w_content}", "style": QType.STMT, "success": True}
+            except:
+                return {"content": "", "style": None, "success": False}
 
 
 async def parse_wh_question(raw_qa_list, pgph_i):
@@ -503,8 +506,10 @@ async def generate_questions(request):
     
     for pgph_i, o in enumerate(__raw_outputs):
         # print(o)
-        print(pgph_i)
+        # print(pgph_i)
     # TODO: Parse ML output to JSON
+        if (not o['success']):
+            continue
         if (o['style'] == QType.WH):
             
             raw_qa_list = []
@@ -527,8 +532,10 @@ async def generate_questions(request):
             __parsed_outputs.extend(await parse_stmt_question(stmts, pgph_i, request.path_params.get('lang', lang.VI_VN)))
     
     # Return the question data
-
-    return JSONResponse({"questions": __parsed_outputs, "paragraphs": __ps, "title": __cont[0]})
+    if len(__parsed_outputs):
+        return JSONResponse({"questions": __parsed_outputs, "paragraphs": __ps, "title": __cont[0]})
+    else:
+        raise HTTPException(500)
 
 async def scan2OCR(request):