Model updates

Official news, first rumor reports and substantive progress. Each entry keeps the information published at that time.

• Stanford's Hazy Research releases NCM, a non-autoregressive language model it says decodes up to 31x faster at matching quality

  Stanford's Hazy Research released NCM with a paper, code and blog post. The researchers report generation 31x faster than an autoregressive baseline on XSum at matching ROUGE-L, 21x faster on LM1B and 10x faster on OpenWebText. On Sudoku, they say NCM beats both autoregressive and diffusion models while using fewer function evaluations.

  OfficialOfficial updateOct 12Source: Hazy ResearchPart of NCM

RSS for this view →