●Change8

v0.1.471-beta

📦 unslothView on GitHub →
✨ 23 features🐛 10 fixes🔧 4 symbols

Summary

This release introduces GLM-5.2 support and a new context length algorithm achieving 3x longer contexts in Unsloth Studio. It also brings significant UI/UX improvements to the Chat Canvas, a redesigned Hub, enhanced security features like `--secure` mode, and better backend stability and hardware support.

Migration Steps

  1. Use `unsloth studio --secure` to enable secure HTTPS global access via Cloudflare.
  2. Update Unsloth or Unsloth Studio using the provided installation scripts (curl for MacOS/Linux/WSL, irm for Windows) to ensure version is `2026.6.9` or `v0.1.471-beta`.

✨ New Features

  • Support for GLM-5.2 model in Unsloth Studio.
  • New auto fit algorithm with MTP enabling 3x longer context lengths for chats.
  • Introduction of Bypass permissions mode in Unsloth Studio.
  • Forkable chats functionality added to the chat canvas.
  • Queue-able chats feature allows queuing new prompts while a generation is running.
  • New hub for model discovery with trending feed, search, and custom model paths.
  • Parallel modules support, allowing concurrent use of Export, Chat, Training, and Recipes.
  • HTTPS Cloudflare support added for secure global access via `unsloth studio --secure`.
  • Chat artifacts are now a 'canvas' with inline HTML canvas cards, Code view, and visible DiffusionGemma raw code.
  • Chat search now covers all messages and prioritizes user messages.
  • Hub features README preview in a split-view feed.
  • Downloads now default to the faster Xet transport with automatic HTTP fallback.
  • New 'Load on selection' toggle in the Hub to set load options before loading a model.
  • Google logo displayed for DiffusionGemma and future Gemma derivatives.
  • DeepSeek-OCR and more vision models now load and run successfully.
  • DiffusionGemma now shows the image forming live during denoising with accurate speed stats.
  • New API server monitor in Studio.
  • Faster API calling and reduced latency.
  • Streamlined logs showing throughput and latency, with bloated logs removed.
  • Improved support for Blackwell RTX 50X and 60X GPUs.
  • Installer now auto-repairs broken or CPU-only PyTorch installs and warns on silent CPU fallback.
  • VRAM for the chat model is freed when training starts, only if GPU memory is tight.
  • Studio now uses a recovery ladder if the llama-server hard-crashes at startup.

🐛 Bug Fixes

  • Fixed fast inference speed-ups on the latest vLLM (0.22+).
  • Tensor parallelism (MTP) now recovers on failure instead of crashing.
  • Fixed silent downgrading to CPU instead of using the GPU.
  • Fixed issue where mmproj VRAM was not accounted for in GGUF fit budget.
  • Fixed keeping local GGUF vision on llama-server.
  • Fixed issue where the install script launch could drain the curl | sh script on WSL/dash.
  • Fixed DiffusionGemma shim not running on clean installs by setting UNSLOTH_IS_PRESENT.
  • Fixed Responses tool output content arrays.
  • Fixed training output directory escaping outputs root when models are on another drive.
  • Fixed llama.cpp prebuilt logic to skip updating already-installed same-release fallbacks.

Affected Symbols

DeepSeek-OCRDiffusionGemmallama.cppvLLM (0.22+)