Ai2 releases Olmo-core 3 open MoE training infrastructure
来源:Ai2 / Hugging Face · huggingface.co
更新摘要
Ai2 released Olmo-core 3, a redesigned open training system for large mixture-of-experts models intended to scale toward one trillion parameters. In one benchmark, the expert pool grew from 8 to 128, total parameters from 4.6 billion to 47 billion, and training throughput fell by less than 5 percent.
本页仅索引并客观转述公开信息,产品能力、价格与可用范围以官方原文为准。