How the size of the model is ~275GB ?
#26
by
chakibb
- opened
Hi everyone,
According to the model card, the model should size 70Bx16bits = 140GB but the storage required is 275GB.
Is it a 32bits or a 16bits model parameters ?
Yes, my bet is if you load and then save in 16 bit nothing will be lost and filesize will match llama 2.
Would be nice to see this fixed though.
What a waste of storage and bandwidth !