Unsloth released version 0.1.908‑beta on 2026‑10‑15, adding the ability to run NVIDIA’s Nemotron 3.5 models and Meta’s Muse Spark Open locally. The Bolt variant of Nemotron 3.5 fits on a 24 GB GPU, while the larger Super VL version needs about 64 GB at 1‑bit precision. Meta’s Muse Spark Open runs on a 256 GB GPU at 2‑bit. The update also improves prompt caching for coding agents and adds support for text‑diffusion training.
Why it matters
Developers can now experiment with high‑capacity LLMs on their own hardware without needing cloud services.
