Unsloth Brings Gemma 4 Local Fine-Tuning

Unsloth is doing the unglamorous but valuable work of turning a headline model release into something people can actually run on consumer hardware. 8GB VRAM for Gemma-4-E2B is the real unlock; it moves local fine-tuning from niche workstation territory into worth-trying territory for hobbyists and developers. The bug fixes are not cosmetic: exploding losses and inference failures are the kind of issues that make fine-tuning feel unreliable even when the model is otherwise strong. Support for E4B, 26B-A4B, and 31B gives teams a ladder from cheap experimentation to heavier workloads, though the larger variants still need serious hardware. The Studio UI plus free notebooks lower friction across text, vision, and audio, which is where Unsloth is differentiating by packaging performance work into something usable.