Y
Hacker News
new
|
ask
|
show
|
jobs
by
toxik
466 days ago
We do clamp probabilities to zero. Look into top-p sampling or nucleus sampling.