Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps

Researchers fine-tuned the LFM2.5-350M language model using Group Relative Policy Optimization (GRPO) in just 100 training steps, improving structured output compliance on the IFStruct benchmark from 22.6% to 29.7%. This lightweight fine-tuning can be done on free-tier GPUs and evaluated locally with llama.cpp.