Morning update on text generation (for context see here, here):
I was doing this really hastily last night and didn’t correctly set up the Mirostat sampling algorithm in the notebook I was using. This led to some undesired repetitiousness.
After fixing this, the samples for @etirabys also look less inaccurate (for example, they use tags like #ramb, which I remember happening in research samples). I may go back and generate the others again too.
the list so far
Will update in reblogs.
Pre-bugfix: voxette-vk, femmenietzsche, xhxhxhx, etirabys
Post-bugfix:
the list so far
Will update in reblogs.
Pre-bugfix: voxette-vk, femmenietzsche, xhxhxhx, etirabys
Post-bugfix:
the list so far
Will update in reblogs.
Pre-bugfix: voxette-vk, femmenietzsche, xhxhxhx, etirabys
Post-bugfix:
(Went back and re-generated @voxette-vk, @femmenietzsche and @xhxhxhx after fixing the aforementioned bug. Dunno if it helped with quality, but my completist neurosis demanded it be done)
the list so far
Will update in reblogs.
Pre-bugfix: voxette-vk, femmenietzsche, xhxhxhx, etirabys
Post-bugfix:
the list so far
Will update in reblogs.
Pre-bugfix: voxette-vk, femmenietzsche, xhxhxhx, etirabys
Post-bugfix:
(via nostalgebraist)
