From da2d8b8267637d041e2557c0e772a2eee6b58a6f Mon Sep 17 00:00:00 2001
From: Kevin Hu <kevinhu.sh@gmail.com>
Date: Tue, 13 Aug 2024 11:21:30 +0800
Subject: [PATCH] boost paralal of graphrag (#1924)

### What problem does this PR solve?


### Type of change

- [x] Performance Improvement
---
 graphrag/index.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/graphrag/index.py b/graphrag/index.py
index ea9c354ad..f2835cc7a 100644
--- a/graphrag/index.py
+++ b/graphrag/index.py
@@ -61,11 +61,11 @@ def build_knowlege_graph_chunks(tenant_id: str, chunks: List[str], callback, ent
 
     assert left_token_count > 0, f"The LLM context length({llm_bdl.max_length}) is smaller than prompt({ext.prompt_token_count})"
 
-    BATCH_SIZE=1
+    BATCH_SIZE=4
     texts, graphs = [], []
     cnt = 0
     threads = []
-    exe = ThreadPoolExecutor(max_workers=12)
+    exe = ThreadPoolExecutor(max_workers=50)
     for i in range(len(chunks)):
         tkn_cnt = num_tokens_from_string(chunks[i])
         if cnt+tkn_cnt >= left_token_count and texts: