quantization? fine-tune?
#68
by
tcporco
- opened
Did anyone make a quantized version of this, or fine tune it? Has anyone ever run it with anything other than 8 A100s?
exactly what i was wondering. If not, i dont understand how this differs from just using the chat bot. Its just an extension of the chat bot as an API to charge
exactly what i was wondering. If not, i dont understand how this differs from just using the chat bot. Its just an extension of the chat bot as an API to charge
open up some of the files and you'll see, it uses reward learning. the chat bot can't be modified as easily. you have to exploit work arounds. yes the hardware demand sucks but its going to take time to bring things to the casual user. also yes an automated lora trainer would be nice.