Method kl_divergence

tokenizer.py:446–450

(self, mu, logvar)

        return mu + eps * std

    def kl_divergence(self, mu, logvar):
        return T.mean(-0.5 * T.sum(
                1 + logvar - mu.pow(2) - logvar.exp(),
                dim=(1, 2))
            )
        
    def repr_from_latent(self, latent: Union[dict, T.Tensor]):
        if isinstance(latent, T.Tensor):