Master Local Fine-Tuning with "gemma-trainer"

Short summary

gemma-trainer is a new skill for fine-tuning Google's Gemma models on local hardware without complex setups. It guides you through Supervised Fine-Tuning, Direct Preference Optimization, and Reward Modeling—with support for multimodal training (text, images, audio) and conversion to lightweight formats (GGUF) for mobile and IoT deployment. The agent-assisted skill includes data validation, LoRA optimization, and comprehensive training reports.

• •Simplifies local fine-tuning of Gemma models with guided workflows for SFT, DPO, and RM
• •Supports multimodal training (text, images, audio) and edge deployment via GGUF to mobile/IoT
• •Agent-assisted with data validation, LoRA parameter tuning, and detailed performance reporting

Generated with AI, which can make mistakes.