●Change8

v0.1.525-beta

📦 unslothView on GitHub →
✨ 12 features🐛 33 fixes🔧 3 symbols

Summary

This release introduces local support for Kimi K3 and DeepSeek v4 Flash models via Unsloth Dynamic GGUFs, along with significant improvements in download efficiency and installation robustness across various systems and operating systems. It also includes numerous bug fixes and enhancements to the Unsloth Studio and desktop applications.

Migration Steps

  1. Update Unsloth using the provided installation commands for macOS, Linux, WSL, or Windows.

✨ New Features

  • Kimi K3 and DeepSeek v4 Flash models can now run locally with Unsloth Dynamic GGUFs.
  • More efficient and faster downloading for Colab, low memory, high memory, and CPU systems.
  • Automatic fallback to HTTP if XET is stuck.
  • DeepSeek V4 Dynamic GGUFs are now supported, with automatic detection of multi-GPU setups and model layer offloading to system memory.
  • DSpark support is now default, providing 2x faster inference.
  • Kimi K3, a 2.8T-parameter MoE model with native vision support and a 1M context window, is now supported.
  • Unsloth supports low, high, and max reasoning efforts for Kimi K3.
  • Unsloth automatically detects multi-GPU setups and offloads layers automatically for Kimi K3.
  • Pre-built Tauri icon for macOS .app bundle in installation.
  • Active training status reporting in Studio.
  • Saved per-model settings applied on API loads, and API monitor added in Studio.
  • Offline mode now detects unreachable hubs, not just dead DNS.

🐛 Bug Fixes

  • Many bug fixes and Mac fixes for smoother installations.
  • Fix two pre-existing Backend CI failures.
  • Warn loudly when requested packing is silently disabled.
  • Preserve reasoning across turns for Qwen3.x.
  • Make tests/utils runnable again and cover test_packing.py in CI.
  • Uninstaller rejects unrecognized arguments instead of uninstalling.
  • Make the uninstaller test suites run on macOS and Windows PowerShell 5.1.
  • Studio: Don't disarm Ctrl+C in agents launched by unsloth start.
  • Studio: Keep a slow model load alive through a proxy timeout.
  • Tighten the comments around the uninstaller argument guard.
  • Route explicit Vulkan setup failures through the setup failure helpers.
  • Honor an explicit map_eos_token=False in get_chat_template.
  • Fix desktop repair version skew and output decoding.
  • Desktop: Improve the installation message.
  • Studio: Make the API keys list scrollable.
  • Desktop: Notarize the final macOS disk image on release.
  • Desktop: Preserve first app window size after setup.
  • data-recipe: Recognize current sampling progress log line.
  • Desktop: Clean up the backend on Unix termination signals.
  • Desktop: Show the app version in Settings > About.
  • Studio: Fix hub search input overlapping the format filter at narrow widths.
  • Desktop: Exit the backend when the app that spawned it dies.
  • Desktop: Open uploaded files via the OS instead of window.open.
  • Run OpenClaw in the launched project workspace.
  • Desktop: Drop documents into a chat instead of rejecting them as non-GGUF.
  • Studio: Fix GGUF classification for custom MTP scan roots.
  • Studio: Complete translation coverage for all supported locales.
  • Guard VLM detection against config.architectures being None in the save paths.
  • Fix multiline prompts through Windows npm shims.
  • Replace deprecated transformers attention mask imports.
  • Assert the Vulkan installer bailouts fail closed via the setup failure helper.
  • Studio: Charge the split compute-buffer rate on a multi-GPU auto fit.
  • Make the auto-appended EOS token deterministic.

Affected Symbols

transformers attention mask importsunsloth-cli.pyconfig.architectures