Kevin Hu
Kevin Hu
Not any updates yet.
I think latest version is OK.
What about increasing the threashold of similarity or decreasing the topN?
What is the chunking method you refer to? If you're not using 'One', you can control chunk size limitation by task page size here:  It limits the page size...
> A PDF file is parsed and splited to chunks, each chunk's size should be limited. I see and agree.
I don't have clue either. onnxruntime seems not to be friendly enough to CUDA things.
We've fixed this in the dev version.
You could use Nginx upstream to full fill this feature.
If the files uploaded in file manager, they're can't be removed in knowledge base.
We didn't try it on the machine of ARM since we didn't have one. You can refer to Dockerfile.scratch which includes all the essential steps to install RAGFlow.