etc.) may break.
    *   Code that uses full path for `get_concrete_function` to trace Keras
        symbolic inputs directly should switch to building matching
        `tf.TensorSpec`s directly and tracing the `TensorSpec` objects.
    *   Code that relies on the exact number and names of the op layers that
        TensorFlow operations were converted into may have changed.

from FIPS 204, passing in ρ, s, and r instead of ρ'. func sampleNTT(rho []byte, s, r byte) nttElement { G := sha3.NewShake128() G.Write(rho) G.Write([]byte{s, r}) var a nttElement var j int // index into a var buf [168]byte // buffered reads from B, matching the rate of SHAKE-128 off := len(buf) // index into buf, starts in a "buffer fully consumed" state for j < n { if off >= len(buf) { G.Read(buf[:]) off = 0 } v := uint32(buf[off]) | uint32(buf[off+1])<<8 | uint32(buf[off+2])<<16 off += 3 f, err :=...