This is a preview of our 8B LLM models finetuned on customer data.
The models are quantized on 4bit (reducing the memory need from 48GB to ~6.5GB) and the last 2 layers are finetuned on
1. wikipedia-tech database subset. – https://34a1665ab6bff5c087.gradio.live/
2. romanian language subset : – https://e80f23c4d37e25859e.gradio.live/
We are limiting to 2 active sessions/model right now. If the links are unavailable please contact us at contact@airl.ro for a demo account.
If you are interested on how it can be deployed on your company website, contact us.
