Vicuna: An Open-Source Chatbot Impressing GPT-4

lmsys.org

44 points·kordlessagain··7 comments

This is nice. But fine-tuning a small model to match a big model and then using the large model to compare the results is troublesome. The same underlying data going round and round ought to result in something like overfitting. Does that happen?

The teacher writes the test to evaluate the students they just taught.

Who else is better equipped to evaluate the students learning?

Some other random teacher, unfamiliar with what has been taught?

Discussion from a month ago: https://news.ycombinator.com/item?id=35378683

The most impressive thing I found in this article was the demo page, quite a few models hosted there. Nice share!

Vicuna is great but not able to be used commercially.

appreciate this comment - saved me from wasting too much time -

I see it mentions ' Terms of Use of the data generated by OpenAI, and Privacy Practices of ShareGPT' on top of the other limits imposed by the project creators.

I wonder if there are any commercially usable, self hostable types of things similar, this one mentions it can be trained for around $300 - I could find uses for something like this..

but I also want something that does not have terms issues from openai or similar as well.

There are multiple models that can be run locally and be used commercially.

https://gpt4all.io/index.html is one provider - CPU inference too!

Reply on news.ycombinator.com