Tired of fragmented mobile AI? Meet M4, the pioneering OS-hardware co-managed foundation model acting like firmware for your smartphone! This LLM-powered solution tackles 38+ diverse tasks across vision, language, audio, and multimodal inputs with accuracy parity in 85% of cases. M4 dramatically simplifies NPU design and boosts on-device intelligence and data privacy, overcoming the "ad-hoc" mess with lightweight "adapters" and significantly fewer operators. Get ready for mobile AI that finally makes sense!

Unlock the full potential of state-of-the-art multilingual LLMs with FineWeb2, a groundbreaking 20 terabyte (5 billion document) dataset. This new pre-training data is generated by a revolutionary curation pipeline that automatically adapts to support any language. Overcoming the inherent difficulty of tailoring filtering and deduplication for a large number of languages, FineWeb2 has been scaled to over 1000 languages using Common Crawl snapshots. It produces more performant models than prior datasets for non-English corpora and includes a principled approach to rebalance datasets for additional performance uplift. Access the released dataset, pipeline, training, and evaluation codebases today!