Negative prompting from the Muse paper?

[LinkAnJarad]

Would it be possible to implement negative prompting as described in section 2.7 in the paper? In the paper, the final tokens are calculated as $\ell_g = (1+t)\ell_c-t\ell_u$, and negative prompting would is done "by replacing the uncondtional logit $\ell_u$ with a logit conditioned on a negative prompt."

[lucidrains]

yea sure! let me see..

[lucidrains]

@LinkAnJarad ok done