Historique des commits

Auteur SHA1 Message Date
  AlpinDale a4cbcfe59f feat: disable logprob serialization to CPU for spec decode il y a 5 mois
  AlpinDale 3a53ff1e01 fix: raise an error for no draft token case when draft_tp>1 il y a 5 mois
  AlpinDale 2c653a2268 fix: make speculative decoding work with per-request seed il y a 5 mois
  AlpinDale 16dff9babc chore: enable bonus token in spec decoding for KV cache based models il y a 5 mois
  AlpinDale d9f4c36edd feat: Medusa speculative decoding support (#590) il y a 5 mois
  AlpinDale dd378ea063 feat: MLPSpeculator with tensor parallel il y a 6 mois
  AlpinDale 7253e9052d feat: integrate typical acceptance sampling for spec decoding il y a 6 mois
  AlpinDale cdff8e89f9 feat: introduce `DraftModelRunner` il y a 6 mois
  AlpinDale b6ff0623a6 chore: clean up branding il y a 6 mois
  AlpinDale abbb730607 feat: support draft model on different tensor parallel size il y a 6 mois
  AlpinDale af43576da0 feat: add MLPSpeculator speculative decoding support (#572) il y a 6 mois
  AlpinDale 313e6e1ec7 feat: add typical acceptance sampling il y a 6 mois
  AlpinDale 4d1e613804 chore: minor simplifications il y a 6 mois
  AlpinDale e0886ee929 feat: add `ProposerWorkerBase` abstract class il y a 6 mois
  AlpinDale 344ddaac5a properly disable speculative decoding il y a 6 mois
  AlpinDale 5b0c11d190 support pipeline parallel pynccl groups il y a 6 mois
  AlpinDale de62ceb18c refactor: eliminate parallel worker per-step task scheduling overhead il y a 6 mois
  AlpinDale 236be273e5 feat: tensor parallel speculative decoding (#554) il y a 6 mois
  AlpinDale 197a6d2c16 auto disable speculative decoding by the running queue size il y a 6 mois
  AlpinDale 438f5bdce9 fix ngrams il y a 6 mois
  AlpinDale ef733aee43 implement ExecuteModelData to reduce executor complexity il y a 6 mois
  AlpinDale 79901b76de logprobs for target model (spec decoding) il y a 6 mois
  AlpinDale 723c6acb84 re-add ngram speculative decoding il y a 6 mois
  AlpinDale 7bcf4c3fc9 centralize gpu worker construction il y a 6 mois
  AlpinDale 9d81716bfd [v0.5.3] Release Candidate (#388) il y a 9 mois
  AlpinDale f8dfac6372 chore: attention refactor and upstream sync apr01 (#365) il y a 10 mois