Commit Verlauf

Autor SHA1 Nachricht Datum
  AlpinDale ae04f57ec1 feat: Pipeline Parallel support (#581) vor 7 Monaten
  AlpinDale 405bb74612 Control plane comms refactor (#573) vor 7 Monaten
  AlpinDale 2c321ce1f2 chore: upgrade to rocm 6.1, update docker vor 7 Monaten
  AlpinDale 323fe23b21 chore: use 127.0.0.1 for single-node setups vor 7 Monaten
  AlpinDale a89c9a0e92 fix: device ordinal issues with world_size and stuff vor 7 Monaten
  AlpinDale 427ab15434 fix: check_health when world_size==1 vor 7 Monaten
  AlpinDale 05d6e43244 fix: `torch.compile()` with mp executor backend vor 7 Monaten
  AlpinDale 5b0c11d190 support pipeline parallel pynccl groups vor 7 Monaten
  AlpinDale de62ceb18c refactor: eliminate parallel worker per-step task scheduling overhead vor 7 Monaten
  AlpinDale 236be273e5 feat: tensor parallel speculative decoding (#554) vor 7 Monaten
  AlpinDale c6a501f682 add multiprocessing executor; make ray optional vor 7 Monaten