All pull requests

Pull requests

• chore: add issue forms, PR template, and CONTRIBUTING guide

  #137·shuk777 opened Oct 15, 2026Collaborator·

  ·

• Arena/ced5a41f edge0 rdna

  #133·Maxritz opened Oct 8, 2026·

  ·

• docs: name peak memory as the MLX allocator peak

  #132·kevin9327 opened Oct 5, 2026·2 tasks done·

  ·
  1

• fix(strip_vision_weights): shrink metadata.total_size by dropped bytes

  #121·kuishou68 opened Sep 27, 2026·

  ·
  1

• docs(streaming): clarify warm_willneed call paths

  #120·agourakis82 opened Sep 23, 2026Contributor·

  ·
  3

• docs(streaming): correct load_hot_layer's "page cache" backing claim

  #118·agourakis82 opened Sep 22, 2026Contributor·

  ·

• fix(scripts): recompute metadata.total_size after stripping vision weights

  #117·agourakis82 opened Sep 22, 2026Contributor·

  ·
  1

  1

• fix(scripts): upload README.md in upload_hf.py

  #116·agourakis82 opened Sep 22, 2026Contributor·

  ·
  1

  2

• tests: make the MLX boundary guard catch mlx_lm imports

  #108·VKVstudio opened Sep 18, 2026·

  ·

• fix(server): decode streamed SSE text from the cumulative id list

  #105·ahrazzle opened Sep 15, 2026Contributor·

  ·

• Draft: local GGUF support with packed Metal execution — feedback on direction

  #23·owenqwenstarsky opened Sep 12, 2026·

  ·
  1

• Add opt-in persistent conversation checkpoints

  #21·owenqwenstarsky opened Sep 12, 2026·

  ·
  2

• backends/cuda: a torch reference backend — edge0-8b and edge0-35b generate text on an NVIDIA GPU

  #19·agourakis82 opened Sep 12, 2026Contributor·

  ·
  21

• fix(streaming): stop layer discovery on IndexError too

  #13·agourakis82 opened Sep 11, 2026Contributor·

  ·
  4

• Harden edge0 serve: loopback bind warning and chat request limits

  #4·inputdrive opened Sep 10, 2026·

  ·
  4