NVIDIA推出Nemotron 3 Nano Omni模型,多模态推理效率提升9倍

Odaily星球日报讯 NVIDIA 在 X 平台发文表示,今日推出开源多模态模型 Nemotron 3 Nano Omni。该模型采用 30B-A3B 混合专家架构(MoE),支持 256K 上下文,可统一处理视频、音频、图像及文本输入。与同类交互水平的开源全能模型相比,该模型吞吐量提升达 9 倍,大幅降低推理成本并提高扩展性。目前 Nemotron 3 Nano Omni 已在 Hugging Face、OpenRouter 及 NVIDIA NIM 上线,并获得 Aible、Applied Scientific Intelligence、H Company 等企业采用。
Disclaimer: OKX Orbit content is provided for informational purposes only. Learn more
Replies
Related Flash News
The governor of South Carolina signed Bill S.163 to protect Bitcoin custody rights and ban discriminatory taxation
India plans to send oil tankers to cross the Strait of Hormuz to restore energy imports to the Middle East
OKX will stop trading USDC in the USDT/USD(S) pair
Two addresses simultaneously opened 10x PEPE long positions, with a total position of $3.37 million
Alibaba's Qianwen model Qwen 3.7-MaX was released
a16z became HYPE's sixth largest holding entity, with a floating profit of $79.29 million
Lighter completes the L2 escape mechanism verification, and users can still withdraw money to the mainnet when the sequencer fails
Multiple Bankr user wallets were stolen, involving Grok's interaction with Bankrbot resulting in unauthorized transaction signatures
Well-known trader Loracle: "HYPE Short Positions TOP 1": Well-known trader Loracle opened 874,155 new long positions of LIT on HyperLiquid
Grafana: The investigation found that recent security incidents have not affected customers' production systems or operations


