How it works: Applies a fixed-strength bias at every generation step. The Problem: In grammatically constrained contexts, forcing a watermarked token can generate unnatural words, degrading text ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback