search

Shark-NLP / DiffuSeq

[ICLR'23] DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models
MIT License
711 stars 87 forks source link

q<n> metrics #45

Closed mainpyp closed 1 year ago

mainpyp commented 1 year ago

When it comes to training metrics, I cannot find the difference between the different q splits. (loss_q1, nll_q0, etc.) Could you shortly explain or reference the corresponding paper / paragraph?

summmeer commented 1 year ago

Hi, We split the metrics according to different timesteps, and thus it is convenient for us to trace the metrics in different timesteps. More specifically, for example, the 4 splits for loss_q stand for the average loss for timesteps [0, 500), [500, 1000), [1000, 1500), [1500, 2000] respectively.