ConceptGenerative Perplexity

Generative Perplexity

A metric for evaluating the quality of generated text; lower values indicate that the model's predictions are more confident and the text quality is higher.

Timeline (last 90 days)

Oct 22

Under a matched NFE budget, recurrent UDLM's generation perplexity is up to 9.8% lower than that of the non-recurrent version.

Unverified50%

All Facts (1)

Unverified

Under a matched NFE budget, recurrent UDLM's generation perplexity is up to 9.8% lower than that of the non-recurrent version.

50%

Source Articles

Looped Uniform Diffusion Language Models: 3.3x Reduction in Training FLOPsOct 22