Kevin Hu
commited on
Commit
·
2118d99
1
Parent(s):
91751af
boost paralal of graphrag (#1924)
Browse files### What problem does this PR solve?
### Type of change
- [x] Performance Improvement
- graphrag/index.py +2 -2
graphrag/index.py
CHANGED
|
@@ -61,11 +61,11 @@ def build_knowlege_graph_chunks(tenant_id: str, chunks: List[str], callback, ent
|
|
| 61 |
|
| 62 |
assert left_token_count > 0, f"The LLM context length({llm_bdl.max_length}) is smaller than prompt({ext.prompt_token_count})"
|
| 63 |
|
| 64 |
-
BATCH_SIZE=
|
| 65 |
texts, graphs = [], []
|
| 66 |
cnt = 0
|
| 67 |
threads = []
|
| 68 |
-
exe = ThreadPoolExecutor(max_workers=
|
| 69 |
for i in range(len(chunks)):
|
| 70 |
tkn_cnt = num_tokens_from_string(chunks[i])
|
| 71 |
if cnt+tkn_cnt >= left_token_count and texts:
|
|
|
|
| 61 |
|
| 62 |
assert left_token_count > 0, f"The LLM context length({llm_bdl.max_length}) is smaller than prompt({ext.prompt_token_count})"
|
| 63 |
|
| 64 |
+
BATCH_SIZE=4
|
| 65 |
texts, graphs = [], []
|
| 66 |
cnt = 0
|
| 67 |
threads = []
|
| 68 |
+
exe = ThreadPoolExecutor(max_workers=50)
|
| 69 |
for i in range(len(chunks)):
|
| 70 |
tkn_cnt = num_tokens_from_string(chunks[i])
|
| 71 |
if cnt+tkn_cnt >= left_token_count and texts:
|