All pull requests

Pull requests

• Studio: preview per-GPU memory before loading

  #13451·Yudeeswaran opened Oct 12, 2026Contributor·

  ·

• Studio: offer a download for files edit_file creates

  #13448·danielhanchen opened Oct 12, 2026Member·

  ·
  9

  

• Studio: serve HTTPS from your own certificate (--ssl-certfile / --ssl-keyfile)

  #13447·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  21

  

• Studio Desktop: choose where Unsloth installs on Windows

  #13446·danielhanchen opened Oct 12, 2026Member·

  ·
  3

  2

  

• Studio: optional info stamp on image downloads

  #13444·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  11

  

• Studio: transcribe with downloaded MLX speech-to-text models on Apple Silicon

  #13443·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  2

  

• Studio setup: build the frontend in the background on source installs

  #13442·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  2

  

• Studio: open a project source from the Sources tab

  #13441·danielhanchen opened Oct 12, 2026Member·

  ·
  6

  

• Fall back to Unsloth inference for Nemotron-H with fast_inference=True

  #13440·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  10

  

• Add unsloth start vscode: put the served model in VS Code's Copilot Chat model picker

  #13439·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  19

  

• fix: release the device cache through the active accelerator on the vLLM resume/teardown paths

  #13438·li-lizhe opened Oct 12, 2026Contributor·

  ·

• studio: restore file drops and composer drag feedback (duplicate of #13387)

  #13436·shimmyshimmer opened Oct 12, 2026Member·

  ·
  5

• Studio: add --models-dir to unsloth studio for local model folders

  #13435·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  14

  

• test_packing_sdpa: take the loss path it reads labels from

  #13434·danielhanchen opened Oct 12, 2026Member·

  ·
  8

• Studio: stop LTX-2 loads crashing Windows with an access violation on 32 GB machines

  #13432·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  8

  

• Studio: browser use follow-ups (safetensors nudge, date fields, disabled dropdowns, moved pages)

  #13429·danielhanchen opened Oct 12, 2026Member·

  ·
  32

  

• Studio: keep AMD integrated GPUs on ROCm on Windows, Vulkan prefill is far slower there

  #13428·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  4

  

• enhance(studio): allow VRAM budget floor of 0% for multi-GPU spreading

  #13427·Shooter3k opened Oct 12, 2026·

  ·

• Studio: composer accepts every document type chat with files parses

  #13426·shimmyshimmer opened Oct 12, 2026Member·

  ·
  36

• Fit Gemma 4 E2B on 8 GB cards on Windows and in Studio chat

  #13425·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  4

  

• Studio: keep Spark-TTS LoRA checkpoints exportable

  #13424·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  17

  

• Studio (AMD iGPU): split an iGPU row's total into VRAM and shared memory

  #13421·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  6

  

• Studio: load sidebar chats, projects and training runs 20 at a time from the server

  #13420·danielhanchen opened Oct 12, 2026Member·

  ·
  32

  

• Studio installer (AMD, Linux): no generic ROCm torch for RDNA 1 / Polaris when the arch is readable

  #13419·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  27

  

• Turn off Granite 3.0-3.2 attention dropout on float16 loads

  #13417·danielhanchen opened Oct 12, 2026Member·

  ·
  1

  11