Closed matpalm closed 1 year ago
apprently this is the expected behaviour of tf.random.normal
!!!
Thanks for the PR. Could you replace the other tf.random.normal call as well?
the noise one? that's actually not being used ( see https://github.com/divamgupta/stable-diffusion-tensorflow/issues/21 ) i'd recommend removing it completely...
oh. oops. pushed that matmul change to the wrong branch. muddies this PR but it's a thing to do anyways to be honest
for whatever reason
tf.random.normal
doesn't honour the passed seed...e.g. consider the following...
this means the current latents generated aren't seeded. switching to numpy, which is fine here since the initial latents aren't in the TF graph at all, fixes things.