Isn't this the small 7B version?
In the microsoft repo the 7B Phi-3 version is called small, medium is the 14B version. That one as 4-bit would be nice too though XP
In the microsoft repo the 7B Phi-3 version is called small, medium is the 14B version. That one as 4-bit would be nice too though XP
Nonono this one is the medium 14B version. Unfortunately we do not support the Small (7B) version.
I guess then huggingface has a bug? Cause it shows in the model card overview 7.36B parameters for me.
EDIT: Just downloaded and tried a training run with this model. It is definitly the 14B version, peft also says so. So the issue is in huggingface ui.
I guess then huggingface has a bug? Cause it shows in the model card overview 7.36B parameters for me.
EDIT: Just downloaded and tried a training run with this model. It is definitly the 14B version, peft also says so. So the issue is in huggingface ui.
Oh weird could you show a screenshot of where it shows that? Thank you so much
@Nazzaroth2 @phamnhuvu it is the phi medium one. the reason it shows 7b is because its the 4bit version = quantized by 50%. the hf numbers arent always accurate