huggingchat/chat-ui · Suggestion - provide API for current models with endpoint for chat (as the huggingchat llama3-70b version is superior to others)

17 days ago

•

Hello,
I've been comparing llama3-70b from huggingchat, groq, deepinfra (via API) and meta.ai and the huggingchat instance is superior to the others.

Is it possible for huggingface to provide a chat API that we can use (with a completion etc) instead of having to run the models ourselves (which appears to be the case at the moment? I have a token key but seems one needs to run the model to use it?).

Alternatively, does anyone know the system instructions that causes the huggingchat instances to give such good responses?

Thanks

SquigglyFruit changed discussion title from Suggestion - provide API for current models (as the huggingchat llama3-70b version is superior to others) to Suggestion - provide API for current models with endpoint for chat (as the huggingchat llama3-70b version is superior to others) 17 days ago

EveryPizza

17 days ago

It might be that HuggingChat has no system instructions by default, and that makes it better. All I know is that meta.ai has a system prompt causing it to act weird, and is much worse than HuggingChat

SquigglyFruit

17 days ago

Sure, maybe.

If we could get access to it via API, paid or not that would be great!

nsarrazin

Hugging Chat org 12 days ago

The models are also available from the API! See the JS docs here for example.

You will need a token from a PRO account for llama 3 70b I think though.

SquigglyFruit

12 days ago

Ah, really? So you don't have to run anything?

That's good, though I was looking for typical API pricing (per token rather than monthly).