Reka AI 发布 190 亿参数全能模型 Rho-1，统一处理文本、图像、视频与机器人控制
Reka AI's omni-model Rho-1 handles text, images, video, and robot control in a single model
Reka AI 发布研究预览版 Rho-1，这是一个 190 亿参数的全能模型，在单一神经网络中处理并生成文本、图像、视频和机器人控制动作。与多数把任务分派给专用模型的系统不同，Rho-1 将所有模态作为 token 放在同一个共享上下文窗口中运行，不调用工具或外部模型，可实时生成连续视频并即时响应新指令而无需重启。预测摄像头图像的同一套权重也驱动机器人动作，为绕开机器人训练数据稀缺，Reka AI 构建了逆动力学模型，从普通互联网视频中提取控制信号；Rho-1 在 320 块 H100 GPU 上训练了约三个月。
正文
当前显示原文，中文全文译文尚未就绪。
Reka AI has released a research preview of Rho-1. The 19-billion-parameter omni-model processes and generates text, images, video, and robot control actions in a single neural network. Unlike most AI systems that route tasks to specialized models, Rho-1 runs all modalities as tokens in one shared context window with no tool calls or external models. The model generates continuous video in real time and responds to new instructions on the fly without restarting.
The same weights that predict camera images also drive robot movements. To work around scarce robot training data, Reka AI built an inverse dynamics model that pulls control signals from ordinary internet videos. Rho-1 trained on 320 H100 GPUs over about three months.
Reka AI isn't new to multimodal AI. In April 2024, the company shipped Reka Core , a multimodal language model that competed with GPT-4, Claude 3, and Gemini Ultra on benchmarks. The release fits a broader push in AI research toward so-called world models .