BF16 of FL32?
#2
by
giannisan
- opened
Hi, I saw that the mergkit configuration says the the type is bfloat26, but the huggingface page says the model is fl32, which one is the correct one?
I used the config to create the model (specifying BF16) but it generated a FP32 model.
Ahh thanks, I'm wondering how this works because i did a merge with mergekit also specifying BF16 but it produces Float16 models when uploaded to huggingface.