I Tested NVIDIA's 30B Nemotron Nano Omni on 18 Tasks — It Killed GPT-4V at 9× the Throughput
中文摘要
NVIDIA开源30B模型，在18项任务中超越GPT-4V，速度是GPT-4V的9倍。
English Summary
NVIDIA’s new 30B Nemotron multimodal model outperforms GPT-4V across key benchmarks while delivering nine times the throughput, all while running efficiently on a single 25GB GPU.
Original Excerpt
NVIDIA quietly open-sourced a 30B multimodal model on April 28 that fits on a single 25GB GPU and tops six leaderboards for document… Continue reading on Medium »