AI discussion! I set up Tavern AI (UI) with Kobold AI (engine) and Pygmalion AI

Wellwisher

Anyone tried setting up Tavern AI (UI) with Kobold AI (engine) and Pygmalion AI? I installed it a few days ago however I haven't managed to explore it as much as ChatGPT.

When ChatGPT came out, I hyper-focused on that for a week asking it a bunch of questions. It's a fascinating technology but it does make mistakes.
As such, there's a danger that all its answers will be taken as face-value by users.
Also I believe ChatGPT is some what crippled by it's ethical and moral filters. Hence I began exploring other AI systems
Another issue I have with it is that Microsoft (knowing it makes mistakes) are looking to implement this across their office platforms (word, excel etc)

I was also wondering what others where doing in the AI genre.

luceos

Wellwisher I truly believe that chat gpt 3.5 turbo and future versions of it, or alternatives that are as capable as this ai, are changing the landscape of how we perceive our day-to-day life. I imagine the effect will be similar as when electricity became widely available.

What people are building with it is already ground breaking:

personal assistants
an automated process for cold acquisition to 100.000 people in their niche market with a personalised, value-adding proposal
saving a dogs life after several failed attempts by vets
role play interaction
complete websites

Wellwisher

Today I successfully ran Dalai AI inside Docker - with 7B LLaMA and Alpaca models on my computer.
I can confirm it does not have strong filters for “inappropriate or explicit” content unlike ChatGPT-3 and the responses are as good as I could “curate” on ChatGPT-3.

You could also run bigger models with up to 65 billion parameters that will make ChatGPT-3 look inferior! 😂

I felt the 7 billion parameters for Dalai AI beats Tavern AI (UI) with Kobold and Pygmalion hands down.

My new gaming PC was able to generate a 200 word story within 30 seconds - which is not bad considering it doesn’t have filters. You can also adjust the length of the response.

Wellwisher

Here's an easier alternative to Dalai with ChatGPT4!

https://github.com/nsarrazin/serge

Basically you install docker
then you run the following code in the command line:
docker run -d -v weights:/usr/src/app/weights -v datadb:/data/db/ -p 8008:8008 ghcr.io/nsarrazin/serge:latest
Download the model of your choice. I recommend 13B, 30B, GPT4All
go to localhost:8008 and enjoy!

MarthaSimons

Wellwisher thanks for guthub url 👍

Its realy help me

Wellwisher

Run ANY Open-Source LLM Locally PC (No-Coding Required)

Previously I used Docker's pre-loaded LLM's but recently switched to LM Studio.

For non-coders LM Studio makes easy work of running LLM's locally.
Before you download an LLM via LM studio it also tells you if your PC can run it

Models I am dating at the moment:
I am kinda digging dolphin-2.5-mixtral-8x7b but they recently released dolphin-2.7-mixtral-8x7b.

You can also run multiple LM's at the same time via the GUI

Wellwisher

Pretty much every company right now 😂