Hugging Face reduces distributed AI training bandwidth by 100x

Original: RT @ClementDelangue: The HF science team just made async RL weight sync ~100x cheaper on bandwidth, and you don't need a shared cluster any…

Source: x.com ↗