infiniflow
[Question]: Knowledge graph construction is extremely slow when using LightRAG. Are there any solutions that can accelerate this extraction process?
Describe your problem
Knowledge graph construction is extremely slow when using LightRAG. Are there any solutions that can accelerate this extraction process?
Same issue. I looked at the log of LLM requests and requests spike every 8 hours, in between its flat at zero.
Is there a timeout or something that makes it wait this long?
Here you can see that it did the processing in 15min and then sitting at last stage for hours. Not LLM activity at this point.
Begin at: Tue, 11 Mar 2025 15:38:11 GMT Duration: 23635.30 s Progress: 15:49:23 Task has been received. 15:49:24 Page(1~9): OCR started 15:50:00 Page(1~9): OCR finished (36.50s) 15:50:07 Page(1~9): Layout analysis (6.61s) 15:50:07 Page(1~9): Table analysis (0.08s) 15:50:08 Page(1~9): Text merged (0.79s) 15:50:08 Page(1~9): Page 0~8: Text merging finished 15:50:08 Page(1~9): Generate 17 chunks 15:50:09 Page(1~9): Embedding chunks (0.71s) 15:50:09 Page(1~9): Done (0.13s) 15:50:11 Start RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval). 15:50:11 Task has been received. 15:50:17 Cluster one layer: 17 -> 4 15:50:25 Cluster one layer: 4 -> 3 15:50:29 Cluster one layer: 3 -> 2 15:50:32 Cluster one layer: 2 -> 1 15:50:32 Done (0.10s) 15:50:35 Entities extraction progress 15:50:35 Task has been received. 15:50:58 Entities extraction progress ... 1/27 (19058 tokens) ... 16:06:48 Entities extraction progress ... 27/27 (13517 tokens)
