36Kr (36氪)→ original

Tencent открывает исходный код ключевых технологий AI Infra Hunyuan

Команда Tencent Hunyuan AI Infra представила HPC-Ops, библиотеку операторов для высокопроизводительного вывода LLM. В реальных сценариях, использование HPC-Ops

AI-processed from 36Kr (36氪); edited by Hamidun News
Tencent открывает исходный код ключевых технологий AI Infra Hunyuan
Source: 36Kr (36氪). Collage: Hamidun News.
◐ Listen to article

The Tencent Hunyuan AI Infra team has taken an important step in advancing artificial intelligence infrastructure by introducing HPC-Ops, a library of key operators for high-performance production-level inference of large language models (LLM), and open-sourcing it. This move will certainly attract the attention of developers and researchers working on optimizing the performance of AI systems.

HPC-Ops is a set of optimized computational kernels designed to accelerate the LLM inference process. Inference, or inferencing, is the stage of using a trained model to generate responses or perform tasks based on input data. This stage is often a bottleneck, especially for large models that require significant computational resources.

According to Tencent Hunyuan's statement, in real-world scenarios, the use of HPC-Ops led to a significant increase in QPM (queries per minute) for various models. In particular, for their own Hunyuan model, this metric increased by 30%, and for the DeepSeek model, by 17%. This is a substantial improvement that can significantly enhance the efficiency and scalability of AI applications.

The open-sourcing of HPC-Ops has several important implications. First, it will allow other companies and researchers to use and adapt this library for their own projects, which will contribute to the dissemination of advanced technologies and accelerate innovation in the AI field. Second, it will enable the community to contribute to the development of HPC-Ops, improving its performance and adding new features. Third, it can stimulate competition in the field of AI infrastructure development, leading to more efficient and accessible solutions.

Tencent's decision to open-source HPC-Ops also highlights a growing trend toward collaboration and knowledge sharing in the AI community. Many companies recognize that an open approach can be more effective than closed development, especially in such a rapidly evolving field as artificial intelligence. This enables faster innovation deployment and the creation of more reliable and resilient systems.

In conclusion, the open-sourcing of HPC-Ops by the Tencent Hunyuan AI Infra team is a significant event that can have a considerable impact on the development of artificial intelligence infrastructure. It is a step that promotes collaboration, innovation, and the dissemination of advanced technologies, which will ultimately lead to the creation of more efficient and accessible AI solutions for various industries and users.

ZK
Hamidun News
AI news without noise. Daily editorial selection from 400+ sources. A product by Zhemal Khamidun, Head of AI at Alpina Digital.

Want to stop reading about AI and start using it?

AI News is a curated feed of AI/tech news. Hamidun Academy teaches you to use AI systematically in your work.

What do you think?
Loading comments…