Gemma 4 emerged as the winner among four local AI models I tested on a literary translation, but with a catch.
Four local AI models translated English to Serbian. Gemma 4 E2B was fastest and passed checks, but still invented words and bent grammar.
llama.cpp
Improving Structured Output Generation by Fine-Tuning a 350M-Parameter Model with Just 100 GRPO Steps
This tutorial shows developers how to improve schema compliance in small local language models by fine-tuning LiquidAI's LFM2.5-350M using GRPO with TRL and LoRA. Using ~500 Nemotron structured-output…