Unsloth Studio Adds Gemma 4 Training Support With Bug Fixes — 1.5x Faster at 60% Less VRAM, 62K Stars

GitHub·medium signal

Unsloth (62K stars) shipped Gemma 4 support including bug fixes, updated chat templates, and llama.cpp fixes for improved tool-calling (updated April 11). Trains Gemma 4 ~1.5x faster while using ~60% less VRAM than FA2 setups — the E2B variant needs only 8GB VRAM for local training. Unsloth Studio now provides a web UI for training and running open models locally, with GGUF export for downstream runtimes like Ollama and LM Studio. Covers E2B, E4B, 26B-A4B, and 31B variants.