●Change8

v0.1.3-beta

📦 unslothView on GitHub →
✨ 10 features🐛 20 fixes🔧 8 symbols

Summary

This release significantly improves tool call accuracy and reliability across models, introduces features like custom folders and preliminary multi-GPU support, and resolves numerous stability and installation issues across platforms.

Migration Steps

  1. For Linux, WSL, or Mac, update Studio using: `unsloth studio update`
  2. For Windows native, update Studio using: `irm https://unsloth.ai/install.ps1 | iex`
  3. For Linux, WSL, or Mac reinstalls, use: `curl -fsSL https://unsloth.ai/install.sh | sh`

✨ New Features

  • Tool calls for all models are now +30% to +80% more accurate.
  • Web search now retrieves actual web content instead of just summaries.
  • Maximum number of allowed tool calls increased from 10 to 25.
  • Improved tool call termination logic to reduce looping and repetitions.
  • Added more tool call healing and de-duplication logic to prevent XML leakage in tool calls.
  • Update button is now visible in the UI.
  • Install script styling has been updated.
  • Added custom folders support, allowing users to use GGUFs located in any folder (accessible via Advanced Settings in Chat and Custom Folders).
  • Preliminary Automatic Multi GPU support added for inference and training, automatically allocating GPU resources for large models.
  • Intel Macs should now work out of the box.

🐛 Bug Fixes

  • Fixed timeouts when downloading large models.
  • Fixed Hugging Face rate limiting issues; HF API calls reduced by 90%.
  • Fixed bun installation issues on Windows and improved overall install speed.
  • Fixed LM Studio GGUF loading on native Windows systems without a GPU.
  • Fixed blank page issue on Windows caused by a broken .js MIME type.
  • Fixed Windows setup.ps1 update-flow.
  • Unified Windows installer/setup logging style, verbosity controls, and startup messaging.
  • Preserved GGUF context max setting after applying and refreshing.
  • Fixed editable install scanning of over 6,500 node_modules directories.
  • Avoided UnicodeEncodeError on Windows cp1252 consoles.
  • Fixed bun Windows bin detection.
  • Skipped download progress polling for exported GGUF models.
  • Replaced hard timeout with an inactivity timeout for model loading.
  • Prevented false multimodal warnings during model loading.
  • Fixed ReadMore links in the Studio tour to open in a new tab.
  • Throttled and cached HuggingFace modelInfo API calls.
  • Corrected default weight_decay and learning rate settings in Studio.
  • Added auto-retry mechanism for stalled HF downloads using HF_HUB_DISABLE_XET=1.
  • Fixed export HF model dropdown clearing behavior on enter or click-away.
  • Simplified tool-call de-duplication and replaced html2text with a builtin converter.

Affected Symbols

unsloth studio updateirm https://unsloth.ai/install.ps1 | iexcurl -fsSL https://unsloth.ai/install.sh | shHF APILM Studio GGUF loadingsetup.ps1device_mapHF_HUB_DISABLE_XET=1