Hugging Face and Unsloth announced free credits for fine-tuning models on Hugging Face Jobs, enabling users to train LiquidAI's LFM2.5-1.2B-Instruct through coding agents like Claude Code and Codex. Unsloth provides approximately 2x faster training and 60% less VRAM usage compared to standard methods. The 1.2B parameter model runs under 1GB of memory and is optimized for on-device deployment on CPUs, phones, and laptops.
No score is assigned. Sources and their independence are shown in the citation chain below.