Update config file and Q on float32
#5
by
RonanMcGovern
- opened
I notice that the Llama 7b model is listed as the name in the config file.
Separately, I see that the dtype is float 32 in the config file. Just confirming that's true? I guess Llama 7B was also trained with float32 but then float16 models are pushed to hub like here
the model is trained with mixed precision, so the model file itself is saved as f32.
thanks! I suppose the model name may be fixed in more final versions as training gets to a close
RonanMcGovern
changed discussion status to
closed
thanks! I suppose the model name may be fixed in more final versions as training gets to a close
Yes Definitely !