GGUF version

#1
by Elfrino - opened

This one looks very interesting. GGUF version possible? :)

Hi @Elfrino

Tagging @TheBloke .

By the way, I haven't had a chance to try this yet. Can you share your feedback if you've tried it?

I will do it now

Best prompt template is ChatML? Or the ### System/### User/### Assistant one used by Intel Neural? Or either?

Hi @TheBloke ,

It works on both ChatML and ### System/### User/### Assistant . But I think the best will be ChatML.

Edit:

You already did it, thanks :)

@Elfrino here is the GGUF version:

https://huggingface.co/TheBloke/OpenHermes-2.5-neural-chat-7B-v3-1-7B-GGUF

Great, that's what I went with. Thanks for the model!

Downloading and will be testing soon, thankyou Weyaxi! and thankyou Bloke for the GGUF!

I tested it out and I find it's pretty solid for the most part but feels a little dry and technical in it's expression and lacks some of the more creativity I have seen in other 7b models. (Mind you, at the moment I have only tested it for it's writing and creativity skills). Still though, it's a good model and I will be testing it out further!

This model has a lot of 'It is essential's and 'it is crucial's in it.

I will do it now

Best prompt template is ChatML? Or the ### System/### User/### Assistant one used by Intel Neural? Or either?

I think best template not ChatML because special tokens isn't added here so if you use ChatML format its probably perform will be lower. You can look

https://huggingface.co/Weyaxi/OpenHermes-2.5-neural-chat-7b-v3-1-7B/blob/main/config.json #32000 vocab size not added

https://huggingface.co/Weyaxi/OpenHermes-2.5-neural-chat-7b-v3-1-7B/blob/main/special_tokens_map.json

Agreed. This ins't working in chatml because special tokens and stuff aren't set up

Sign up or log in to comment