Sounds interesting - could you do a model card to explain more what it is?
I regularly get people asking me about non-English fine tuned models. Is this a German language fine tune? If so, I'd love to quantise it and publicise it.
But it'd be great if there was a little info in the model card so people know what it is and how it was made?
Thanks in advance
https://huggingface.co/flozi00/Llama-2-13B-german-assistant-v2
Hi, take that model instead :-)
I was planning adding an 4bit autogptq version too, but if you could do so I can focus my GPU's on training :D
Glad to! Could you do a quick model card for it? Then my code pulls that into my README automatically, and I know what to tell people about it when I announce it
https://huggingface.co/flozi00/falcon-7b-german-assistant-v2
Done, there is also one more model that could be insteresting for you
Thanks! I put the Llama quants out yesterday and I will look at the Falcon one later today.
Unfortunately it seems like the GGML versions of the Llama 2 German Assistant aren't working well. But I've not had any complaints about the GPTQ versions, so hopefully they're working OK for people.
Feel free to ping me any time you release new models if you'd like me to quantise them.