Depedencies Installation

[scalenow]

Description

What is the correct version to install graphrag future,and LightRAG ,Nano-GraphRAG dependencies on running the command pip install graphrag future it is breaking the package.

(documentanalysis) root@scalenowai:~/miniconda3/envs/documentanalysis/kotaemon# python app.py
GraphRAG dependencies not installed. Try pip install graphrag future to install. GraphRAG retriever pipeline will not work properly.
LightRAG dependencies not installed. Try pip install git+https://github.com/HKUDS/LightRAG.git to install. LighthRAG retriever pipeline will not work properly.
Nano-GraphRAG dependencies not installed. Try pip install nano-graphrag to install. Nano-GraphRAG retriever pipeline will not work properly.
No module named 'lightrag'
User "admin" already exists
Setting up quick upload event
Running on local URL: http://127.0.0.1:7860

Reproduction steps

What is the correct version to install graphrag future,and LightRAG ,Nano-GraphRAG dependencies on running the command pip install graphrag future it is breaking the package.

(documentanalysis) root@scalenowai:~/miniconda3/envs/documentanalysis/kotaemon# python app.py
GraphRAG dependencies not installed. Try `pip install graphrag future` to install. GraphRAG retriever pipeline will not work properly.
LightRAG dependencies not installed. Try `pip install git+https://github.com/HKUDS/LightRAG.git` to install. LighthRAG retriever pipeline will not work properly.
Nano-GraphRAG dependencies not installed. Try `pip install nano-graphrag` to install. Nano-GraphRAG retriever pipeline will not work properly.
No module named 'lightrag'
User "admin" already exists
Setting up quick upload event
Running on local URL:  http://127.0.0.1:7860

Screenshots

![DESCRIPTION](LINK.png)

Logs

No response

Browsers

No response

OS

No response

Additional information

No response

[jrista]

I am running into the same problem. Installing graphrag future results in a slough of dependencies breaking. Is there a specific version of graphrag future that is required? For that matter, I was trying to use nano-graphrag and I suspect the same issue exists there, that a specific version is required. It seems any time a rag package is installed, the versioning of various dependencies gets screwed up.

[jrista]

Ok. Utilized Claude a bit to try and help me resolve the issue. You only need to install one of the three RAG options. Pick EITHER nano-graphrag, lightrag, or graphrag+future. For the latter, if you choose that, Claude seemed to determine the following were the proper packages:

pip install "graphrag<=0.3.6" future

I am not 100% certain that <=0.3.6 is indeed the best, I still had a couple of packages that were wrong.

I decided to just go the nano-graphrag route, installed only that, then followed the instructions to uninstall and reinstall the hnswlib package, and that seemed to have worked. NanoGraphRAG seems to be functioning in my local install now. I used coda and nuked my environment entirely, and recreated it, in order to start with a truly fresh python environment, as my previous was too far gone with too many incorrect packages.

[Carb91]

I'm experiencing the same issue on Mac with Nano GRAPHRAG.
I tried installing and uninstalling hnswlib package but the problem wasn't resolved.
I've installed kotaemon via Docker. Have you found any solution?

[jrista]

Just curious, what version of python are you running? I've noticed that with many RAG/AI libs, if you are using a version of py newer than 3.10, then you may run into additional problems. I started explicitly choosing python=3.10 when setting up my conda environments, and that seems to have resolved many problems I've been facing with RAG stuff lately.

Outside of that, make sure you uninstall, then REINSTALL, the hnswlib packages.

[scalenow]

Python 3.10,this API was summarising 40 pages document in 30 seconds now takes 30 minutes .The upgrade has compromised the performanceok of the application

[Carb91]

I' m running Python 3.11

[jrista]

Python 3.10,this API was summarising 40 pages document in 30 seconds now takes 30 minutes .The upgrade has compromised the performanceok of the application

Which upgrade?

[scalenow]

@jrista The older version ,see the pic attached .I dont remember the version no

[jrista]

@jrista The older version ,see the pic attached .I dont remember the version no

Version of Kotaemon, or older version of Python?

[scalenow]

@jrista older version of kotaemon, python version 3.10

[jrista]

@jrista older version of kotaemon, python version 3.10

Ok, thanks. So the previous version of Kotaemon was actually a lot faster... We have a pretty specific need, and I'm starting to wonder if rolling our own thing (using some libs just for the core LLM RAG aspects) might be an easier route. We've been having some trouble integrating with Kotaemon, without at least learning its entire codebase.