top | item 42342973

(no title)

vletal | 1 year ago

That would be 1.38 bits per weight on average, which I can confidently guess would not perform well.

discuss

qeternity|1 year ago

BitNet is functional at 1.58 bpw.

Lerc|1 year ago

The model card says the 70B is 16 bit so I think you have twice that