marella / chatdocs

Chat with your documents offline using AI.
MIT License
707 stars 100 forks source link

Is chatdocs still being supported? #101

Open velazquej1 opened 9 months ago

velazquej1 commented 9 months ago

Just curious if this and ctransformers are still being supported? There has been no activity and no updates in close to 6 Months that I've seen.

BharatBlade commented 9 months ago

Doesn't look like it. The issue I raised last hasn't been addressed. This was a really amazing project that worked seamlessly for a very long time (~1 year). My guess based on the errors I was/am getting, is that huggingface python libraries changed and calls made to it stopped working properly for stuff idk what for (embeddings maybe?). I've been trying to find a seamless solution like this that can be deployed independently for users who aren't programmers. I personally program in Java, not Python, so the nuances with Python are enough to stop me from being able to mod this project currently. Hopefully later I'll get more time to do a deeper dive and figure out what to do to fix it. Suggestions/corrections to what I said are highly welcome.

I do have a somewhat working version of it on my laptop, but only because I haven't updated python packages at all. I set it up forever ago, it worked, and I never did anything python related on my laptop. All the packages haven't changed since I built it. I get a ton of warnings but it still works, with my Nvidia laptop GPU too. I tried migrating my environment with pip freeze (I installed the whole thing globally, not a venv) to my much higher performing desktop but it still failed. If you have any thoughts on how to replicate what I have on my laptop for other devices/users, I would greatly appreciate it. Would be happy to share whatever I had as well in terms of python package files if it helps as well.

e-e-eric commented 4 months ago

Google NotebookLM does what I wanted this to do. It's not offline, but data is private. It's also much quicker than a default setup of PrivateGPT.

For a small amount of docs (<50 per instance) I wouldn't bother self-hosting atm.