Unsloth released v0.1.904-beta, enabling conversion of any large language model—including Qwen3.5, Gemma-4, and Llama—into a decision model styled after Jev, Clef, and Laya1. The project reported classification accuracy rising from approximately 30% to 80% after training. The release also adds native loading of ComfyUI diffusion models, faster diffusion inference via INT8 ConvRot, and platform-specific sandboxing using Bwrap on Linux, Seatbelt on Mac, and MXC on Windows.
Unsloth v0.1.904 Converts Any LLM Into a Decision Model
Unsloth's v0.1.904-beta release enables converting any LLM into a decision model, reporting classification accuracy gains from roughly 30% to 80% after…