Pull requests

• Studio: flag a modules.json folder of only dots and slashes that leaves the model

  #14527·danielhanchen opened Oct 21, 2026Member·

  ·

• Studio: a mermaid style statement after a semicolon cannot fetch either

  #14526·danielhanchen opened Oct 21, 2026Member·

  ·

• Studio: show the AMD NPU by default again, with a clearer setup notice

  #14524·danielhanchen opened Oct 21, 2026Member·

  ·
  8

  

• Studio: load GGUF on CPU when the auto-selected Vulkan driver hangs or has no device

  #14521·danielhanchen opened Oct 21, 2026Member·

  ·
  11

  

• FastLanguageModel: dequantize fine-grained fp8 checkpoints on an explicit load_in_16bit

  #14516·danielhanchen opened Oct 21, 2026Member·

  ·
  7

• Studio: save sampling defaults per model and use them for API requests

  #14515·danielhanchen opened Oct 21, 2026Member·

  ·
  1

  38

  

• Studio: fix host memory checks and improve logging for GGUF image loads

  #14514·VTplus opened Oct 21, 2026Contributor·

  ·
  1

• QAT: qat_scheme = "nvfp4", and strip NVFP4 / MX QAT linears before saving

  #14512·danielhanchen opened Oct 21, 2026Member·

  ·
  22

• Studio: refuse a single-file diffusion load that would exhaust host RAM instead of crashing

  #14510·danielhanchen opened Oct 21, 2026Member·

  ·
  28

  

• Studio Images: paste generation parameters, inpaint mask blur and only-masked area, seed buttons

  #14508·danielhanchen opened Oct 21, 2026Member·

  ·
  18

  

• Studio: move formatRate to the next unit when rounding reaches 1024

  #14505·kevin9327 opened Oct 21, 2026Contributor·

  ·
  1

• Bump the cargo-tauri group in /studio/src-tauri with 2 updates

  dependenciesrust
  #14503·dependabot[bot] opened Oct 21, 2026Bot·

  ·

• Studio: optional reply times in Settings > Chat

  affects UI#edge case#new-feature#>3000#
  #14500·danielhanchen opened Oct 21, 2026Member·

  ·
  1

  29

  

• Studio: extract ebooks, Office files and email in Data Recipes document seeds

  affects UI#new-feature#security#
  #14499·danielhanchen opened Oct 21, 2026Member·

  ·
  1

  25

  

• Studio: OpenAI-compatible fine-tuning jobs API (/v1/files, /v1/fine_tuning/jobs)

  new-feature#
  #14496·danielhanchen opened Oct 21, 2026Member·

  ·
  1

  41

  

• Jetson / Rosetta follow-ups: an explicit GPU index beats the Jetson CPU record, JetPack 7 off the CPU note

  codex-converged#edge case#
  #14494·danielhanchen opened Oct 21, 2026Member·

  ·
  5

• XPU bf16 on the current device, keep explicit bf16 on XPU, MLX full finetuning keeps its default dtype

  edge case#non-studio#
  #14493·danielhanchen opened Oct 21, 2026Member·

  ·
  13

• Studio: discrete-first GGUF placement keeps explicit device picks, matching context ceiling; in-process Rosetta check

  edge case#
  #14492·danielhanchen opened Oct 21, 2026Member·

  ·
  24

• Studio: FLUX.3 upsampler pin follows torch's vendor, single-GPU hosts launch as before, non-blocking Vulkan probe

  edge case#regresses#
  #14491·danielhanchen opened Oct 21, 2026Member·

  ·
  17

• MoE grouped GEMM: share autotuners across identical cards; fix the TMA test on one-GPU hosts and macOS

  codex-converged#non-studio#
  #14490·danielhanchen opened Oct 21, 2026Member·

  ·
  3

• Studio: stop the SDPA probe reporting fused attention on MPS

  codex-converged#
  #14488·Lyxot opened Oct 21, 2026Collaborator·

  ·
  3

  

• Clef: keep fp16 loss scaling on the float32 path (fixes Gemma 3 not training on T4)

  important#non-studio#
  #14482·indrajeetapache opened Oct 21, 2026Contributor·

  ·
  1

  4

  

• Fetch get_statistics repos concurrently so from_pretrained does not wait on four HF round trips

  non-studio#
  #14476·danielhanchen opened Oct 21, 2026Member·

  ·
  1

  7

  

• Fix file drops in new project chats

  affects UI#new-feature#refactor#
  #14472·Imagineer99 opened Oct 21, 2026Member·

  ·
  1

• perf(studio): prefill MLX loads in the planned step while experts are on disk

  affects UI#conflicts#edge case#MLX>3000#
  #14466·Lyxot opened Oct 21, 2026Collaborator·

  ·
  7