Meituan, a Chinese company, has developed LongCat-2.0, a 1.6 trillion parameter AI model. Notably, the model was trained entirely on domestic AI chips, avoiding NVIDIA hardware. The model's efficiency stems from its Mixture-of-Experts design, which utilizes only 3% of its parameters per token, making it feasible to train on less powerful, non-NVIDIA accelerators. AI
IMPACT Demonstrates feasibility of training large models on non-NVIDIA hardware, potentially diversifying the AI infrastructure landscape.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=2 ai=1.0]
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
