Science & Tech · AI · 21 hrs ago

DeepSeek open-sources AI inference framework for Nvidia and Huawei chips

DeepSeek open-sources AI inference framework for Nvidia and Huawei chips

Chinese AI company DeepSeek has open-sourced DeepInfer, a framework it developed to run AI models.

It supports Nvidia GPUs and Huawei Ascend processors, and currently works with DeepSeek's V4.1-Flash model.

Inference is the process of running an AI model to respond to requests, and frameworks can help make that process more efficient.

DeepInfer includes tools for handling large workloads and producing structured outputs.

DeepSeek says the framework can improve speed while cutting costs and energy use, which matter to businesses running AI services.

The company also open-sourced a component for processing API requests and model inputs and outputs.

DeepSeek plans to release more inference code and improvements for future models through the project.

4 sources

Related

What people are saying

Senior researcher at DeepSeekCompany researcher
light and darkness
via SCMP