Change8

v0.1.43-beta

Breaking Changes
📦 unslothView on GitHub →
1 breaking7 features🐛 10 fixes🔧 3 symbols

Summary

This release focuses heavily on improving Unsloth Studio functionality, adding Gemini provider support, and enhancing cross-platform compatibility, especially around `llama.cpp` binaries for various macOS and Windows CUDA versions. Users are strongly advised to use the new installation scripts instead of `unsloth studio update`.

⚠️ Breaking Changes

  • DO NOT USE `unsloth studio update` as packaging will not get the latest updates; use the provided curl/irm scripts instead.

Migration Steps

  1. To update Unsloth or install Unsloth Studio, use the provided installation scripts (curl for macOS/Linux/WSL, irm for Windows) instead of `unsloth studio update`.
  2. If you encounter issues on older macOS versions, note that Apple Silicon Mac OS 13 (Ventura) now requires a source build for `llama.cpp`, while newer versions use prebuilt binaries.

✨ New Features

  • Studio: expose --parallel / -np flag on `unsloth studio run`.
  • Studio: add frontend i18n support.
  • Studio: expose image size setting in training UI.
  • Studio: add configurable CPU thread pool limit.
  • Studio: add Gemini provider with web_search, code_execution, prompt caching, and Nano Banana image generation.
  • Studio: add remote MCP server support.
  • Studio: add stdio MCP server support.

🐛 Bug Fixes

  • Fixed issue where CUDA 13.2 caused gibberish output (resolved by CUDA 13.3 support).
  • Fixed non-streaming GGUF chat completion usage.
  • Fixed Studio Colab proxy and iframe issues preventing loading.
  • Fixed issue where context-resizing caused an error.
  • Fixed issue where x86_64 Python venv on Apple Silicon was not detected and rebuilt as arm64.
  • Fixed Strix halo and added Windows AMD ROCm support.
  • Fixed wrong code block language for nightly Windows install.
  • Fixed phantom scroll area below messages with many sources in Studio.
  • Fixed error when context-resizing.
  • Fixed issue where Studio incorrectly loaded diffusion GGUFs as chat models.

Affected Symbols