Re: [spam] [personal] perceiver model notes

[k]

trains with reducing loss, even on rasbpi, but the outputs don't
differentiate yet due to some problem.  every output byte is the same
value: 
https://ipfs.io/ipfs/bafkreig34njqveooj4yrx3qpyttrvimxyibvtysqegy73yi2e4vfejnqdy