jordiclive

/

gpt4all-alpaca-oa-codealpaca-lora-13b

like 11

I see you combined a few good instruction datasets. I was considering doing the same, but haven't found the time. How are the results? What's your impression.

There seems to be a couple of school of thought, one is that is better to go smaller, better quality, and cleaner data. Another is to combine lower quality datasets to get a large amount of instruction.

It seems like the alpaca datasets is the cleanest right now. So does this get better results that the alpaca_cleaned models?