Alibaba Cloud, the digital backbone of Alibaba Group, made a significant announcement at the Apsara Conference, unveiling over 100 new open-source Qwen 2.5 multimodal models and a groundbreaking text-to-video AI model designed to elevate visual creativity.
100 Open-Sourced Models for Developers
The highlight of the event was the introduction of the Qwen 2.5 models, which range in size from 0.5 to 72 billion parameters. These models boast improved mathematical and coding capabilities and support over 29 languages, making them applicable across diverse sectors, including automotive, gaming, and scientific research. Since their debut in April 2023, the Qwen series has achieved over 40 million downloads, further inspiring the creation of more than 50,000 models within the open-source community.
Jingren Zhou, CTO of Alibaba Cloud Intelligence, emphasized the initiative's potential to empower developers and organizations of all sizes, stating, "Today marks a significant milestone as we launch our most expansive open-source initiative to date."
Advancements in Multimodal AI
In addition to the new language models, Alibaba Cloud introduced a new text-to-video model, part of its Tongyi Wanxiang large model family. This innovative model can generate high-quality videos from text prompts in both Chinese and English, transforming static images into dynamic video content. Featuring advanced diffusion transformer (DiT) architecture, it promises improved video quality and versatility.
The unveiling also included the Qwen2-VL, an advanced vision language model capable of understanding videos up to 20 minutes long and supporting video-based question-answering. This model is designed for integration into mobile devices, vehicles, and robotics, enhancing automation capabilities.
Revamped Infrastructure for AI Computing
To support the increasing demand for robust AI computing, Alibaba Cloud announced a comprehensive upgrade to its full-stack infrastructure. Key highlights include:
- Next-Gen Data Center Architecture: The CUBE DC 5.0 architecture enhances energy efficiency and reduces deployment times by up to 50% with innovative technologies, including a wind-liquid hybrid cooling system.
- Open Lake Solution: This solution integrates big data engines into a unified platform, maximizing data utility for generative AI applications and optimizing resource usage.
- AI Scheduler: A new cloud-native scheduling engine that improves compute resource management, achieving over 90% effective utilization rates through intelligent integration and real-time adjustments.
- Unified Metadata Management: The DMS: OneMeta+OneOps platform streamlines management across various data sources, significantly boosting data utilization rates.
- Enhanced Elastic Compute Service: The 9th Generation ECS instance offers substantial performance improvements, including a 30% increase in search recommendation speed.
Eddie Wu, Chairman and CEO of Alibaba Cloud Intelligence, stated, “Alibaba Cloud is investing with unprecedented intensity in AI technology and infrastructure to unlock our customers’ business potential.”
With these groundbreaking developments, Alibaba Cloud is positioning itself at the forefront of AI innovation, aiming to foster the growth of generative AI technologies across various industries while maximizing value for its global customer base.