Fix: metadata issue & graphrag speeding up. (#12113)

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

---------

Co-authored-by: Liu An <asiro@qq.com>

graphrag/general/extractor.py

@@ -71,18 +71,17 @@ class Extractor:
         _, system_msg = message_fit_in([{"role": "system", "content": system}], int(self._llm.max_length * 0.92))
         response = ""
         for attempt in range(3):
-
             if task_id:
                 if has_canceled(task_id):
                     logging.info(f"Task {task_id} cancelled during entity resolution candidate processing.")
                     raise TaskCanceledException(f"Task {task_id} was cancelled")
-
             try:
                 response = asyncio.run(self._llm.async_chat(system_msg[0]["content"], hist, conf))
                 response = re.sub(r"^.*</think>", "", response, flags=re.DOTALL)
                 if response.find("**ERROR**") >= 0:
                     raise Exception(response)
                 set_llm_cache(self._llm.llm_name, system, response, history, gen_conf)
+                break
             except Exception as e:
                 logging.exception(e)
                 if attempt == 2:

graphrag/general/index.py

@@ -198,7 +198,7 @@ async def run_graphrag_for_kb(
 
         for d in raw_chunks:
             content = d["content_with_weight"]
-            if num_tokens_from_string(current_chunk + content) < 1024:
+            if num_tokens_from_string(current_chunk + content) < 4096:
                 current_chunk += content
             else:
                 if current_chunk:

graphrag/light/graph_extractor.py

@@ -78,10 +78,6 @@ class GraphExtractor(Extractor):
         hint_prompt = self._entity_extract_prompt.format(**self._context_base, input_text=content)
 
         gen_conf = {}
-        final_result = ""
-        glean_result = ""
-        if_loop_result = ""
-        history = []
         logging.info(f"Start processing for {chunk_key}: {content[:25]}...")
         if self.callback:
             self.callback(msg=f"Start processing for {chunk_key}: {content[:25]}...")