<ul><li>Microsoft has released a small language model called Phi-4 that can solve certain math problems better than larger algorithms.</li><li>Phi-4 was trained mostly on synthetic data rather than web content, suggesting that incorporating more synthetic files into small models' training datasets could enhance reasoning skills.</li><li>Phi-4 features an upgraded tokenizer and an enhanced attention mechanism that can analyze 4,000 tokens of user input.</li><li>Phi-4 outperformed its predecessor and other models on math-related reasoning tests, despite having fewer parameters.</li></ul>

Microsoft releases Phi-4 language model trained mainly on synthetic data

Discover more