Unsloth v0.1.905-beta adds code sandboxing and QLoRA decision-model training | Tool Update
Unsloth published release v0.1.905 beta on 8 Oct 2026 at 15:43. The release adds platform specific code sandboxing and QLoRA decision model training with suppo…
Published on MyPrivateClaw
Oct 9, 2026, 5:37 PM UTC
Coverage date
Oct 9, 2026
Last updated
Oct 9, 2026, 5:37 PM UTC
News summary
The v0.1.905 beta release lets users train any text or vision LLM as a Jev style decision model using QLoRA, export Clef models with Qwen3.5 backbones and Laya models to GGUF, and serve supported decision models through llama.cpp. The stated accuracy claim of 30% to 80% is a vendor self report with no independent benchmark. The release introduces sandboxing for the code models run: Bwrap on Linux, Seatbelt on Mac and MXC on Windows, with a protection level selectable in Settings. Unsloth states that QLoRA on A100 and RTX PRO 6000 trains Qwen3.5 35B A3B up to 4.1x faster and Qwen3 30B A3B up to 3.3x faster, and enables INT8 ConvRot for Qwen Image 2.1 diffusion by default. These performance multipliers are unverified vendor figures. Unsloth Desktop is free and open source, distributed for Windows, macOS and Linux via GitHub releases, with GGUF export support. Note: the Unsloth changelog l…