How Tencent’s Angel Platform Secured the 2024 World Internet Conference Leading Technology Award
Tencent’s Angel machine learning platform, recognized for breakthroughs in trillion‑scale model training, inference, and deployment, won the 2024 World Internet Conference Leading Technology Award, highlighting its self‑developed hardware‑software stack, high‑performance networking, and extensive real‑world AI applications.
Tencent Angel machine learning platform recently won the 2024 World Internet Conference Leading Technology Award.
The platform has overcome challenges in trillion‑parameter model distributed training, inference, and large‑scale application deployment, achieving autonomous R&D from hardware to critical software technologies and driving significant advances in industry and the digital economy.
The award, organized by the World Internet Conference, honors the most leading internet technology achievements worldwide and promotes international collaboration.
In total, 20 internationally representative projects were awarded this year, covering brain‑like computing, embodied intelligence, 6G, big data, high‑performance chips, and AI machine‑learning platforms, with an overall acceptance rate of only 7.4%.
Vice President Jiang Jie stated that the award recognizes Tencent’s self‑developed large‑model technologies; with model parameters exceeding a trillion and training data over ten trillion, the Angel platform addresses these massive challenges, delivering notable performance and efficiency gains, and Tencent will continue to invest in foundational technology R&D and open‑source sharing.
Angel’s breakthroughs span three areas: network interconnect, framework efficiency, and platform scale.
In networking, Tencent’s self‑developed RDMA solution provides high‑speed cluster interconnect with per‑node bandwidth of 3.2 TB, delivering a 30% performance boost and 70% cost reduction compared to leading competitors.
In efficiency, a custom training/inference framework with unified memory and multi‑dimensional parallelism, along with operator optimizations, yields 2.6× faster training and 2.3× faster inference versus open‑source frameworks.
In scale, the platform achieves linear scaling up to ten thousand GPUs, raising GPU cluster acceleration from 50% to 99%.
The Angel project has previously topped international rankings, winning over 20 international competition championships.
Angel is also China’s first top‑tier AI open‑source project graduating from the Linux Foundation, named the 2019 most popular Chinese open‑source software, attracting many external developers and enterprise users, fostering broad technology adoption and industry collaboration.
In March 2024, the platform received the 2023 China Institute of Electronics Society’s first‑class Science and Technology Progress Award.
Based on Angel, Tencent’s internally developed mixed‑expert (MoE) model “Hunyuan Large” broke the 389 billion‑parameter barrier with 52 billion active parameters and a context length of 256 K, trained and inferred on the Angel platform.
Currently, Angel supports models ranging from 0.5 B to 70 B in size, covering language models, multimodal understanding, and multimodal generation.
These models have been deployed in over 700 Tencent business scenarios, demonstrating strong application potential; externally, Angel is offered via Tencent Cloud, serving sectors such as advertising, recommendation, social, and finance, and supporting 300 000 industry customers for digital transformation.
Congratulations to all Tencent engineers involved.
Tencent Tech
Tencent's official tech account. Delivering quality technical content to serve developers.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.