Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs

Hugging Face has released Olmo-core 3, an open-source training infrastructure designed specifically for large Mixture-of-Experts models. By providing standardized code for efficient distributed training, this release aims to lower the barrier for developers building complex architectures. The project offers a transparent alternative to proprietary training stacks, potentially accelerating research into large-scale machine learning systems that require high computational throughput.