Commit History

Autor SHA1 Mensaxe Data
  AlpinDale e0886ee929 feat: add `ProposerWorkerBase` abstract class hai 7 meses
  AlpinDale 344ddaac5a properly disable speculative decoding hai 7 meses
  AlpinDale 5b0c11d190 support pipeline parallel pynccl groups hai 7 meses
  AlpinDale de62ceb18c refactor: eliminate parallel worker per-step task scheduling overhead hai 7 meses
  AlpinDale 236be273e5 feat: tensor parallel speculative decoding (#554) hai 7 meses
  AlpinDale 197a6d2c16 auto disable speculative decoding by the running queue size hai 7 meses
  AlpinDale 438f5bdce9 fix ngrams hai 7 meses
  AlpinDale ef733aee43 implement ExecuteModelData to reduce executor complexity hai 8 meses
  AlpinDale 79901b76de logprobs for target model (spec decoding) hai 8 meses
  AlpinDale 723c6acb84 re-add ngram speculative decoding hai 8 meses
  AlpinDale 7bcf4c3fc9 centralize gpu worker construction hai 8 meses
  AlpinDale 9d81716bfd [v0.5.3] Release Candidate (#388) hai 10 meses
  AlpinDale f8dfac6372 chore: attention refactor and upstream sync apr01 (#365) hai 11 meses