Zhipu AI Unveils Open Source GLM-Edge Model Series
date
Dec 1, 2024
damn
language
en
status
Published
type
News
image
https://www.ai-damn.com/1733054496342-202406051435016830_1.jpg
slug
zhipu-ai-unveils-open-source-glm-edge-model-series-1733054505178
tags
GLM-Edge
Zhipu AI
Edge AI
Open Source
Large Language Models
summary
Zhipu Technology has launched the open-source GLM-Edge series, featuring large language and multimodal models optimized for edge devices. The series, consisting of four models, aims to enhance real-world applications on mobile and desktop platforms, showcasing significant advancements in processing speed and efficiency.
Zhipu AI Unveils Open Source GLM-Edge Model Series
Zhipu Technology has announced the open-source release of its edge-side large language and multimodal model series, GLM-Edge. This initiative represents a significant step forward in the company's efforts to implement practical applications on edge devices, catering to the growing demand for efficient AI solutions.
Overview of GLM-Edge Models
The GLM-Edge series comprises four distinct models:
- GLM-Edge-1.5B-Chat
- GLM-Edge-4B-Chat
- GLM-Edge-V-2B
- GLM-Edge-V-5B
These models are optimized for various platforms, including mobile devices such as smartphones and automotive systems, as well as traditional desktop environments like personal computers. This broad compatibility is designed to facilitate the deployment of advanced AI capabilities across multiple use cases.
Technological Advancements
Building upon the technological foundation of the GLM-4 series, Zhipu's research team has restructured and resized the models to achieve an optimal balance between performance, real-time inference capabilities, and deployment ease. The company has engaged in extensive collaboration with partners to optimize inference processes, which has resulted in impressive operational speeds on selected edge platforms.
Particularly notable is the performance on the Qualcomm Snapdragon 8 Elite platform, where the models leverage NPU computing power alongside a mixed quantization approach. The 1.5B chat model and the 2B multimodal model are capable of decoding speeds exceeding 60 tokens per second. Furthermore, with the application of speculative sampling techniques, decoding speeds can surpass 100 tokens per second.
Impact on Edge AI Applications
The open-source nature of the GLM-Edge series not only highlights Zhipu's technological expertise in artificial intelligence but also empowers developers and researchers with robust tools and resources. These resources are intended to drive the growth and innovation of edge AI applications, fostering a more accessible environment for experimentation and development.
Conclusion
With the introduction of the GLM-Edge series, Zhipu Technology is poised to make significant contributions to the field of edge AI. By making these models available to the public, the company aims to encourage a collaborative approach towards the advancement of AI technologies in real-world applications.
GLM-Edge Collection:
Key Points
- Zhipu AI has launched the open-source GLM-Edge model series.
- The series includes four models optimized for various platforms.
- Models demonstrate impressive decoding speeds, enhancing real-time applications.
- Open-source availability promotes innovation in edge AI development.