TAIDE models added to Qualcomm AI Hub: Empowering Trustworthy AI Dialogue

Qualcomm Technologies is proud to announce that TAIDE (Trustworthy AI Dialogue Engine), an open source LLM model, is now available on Qualcomm AI Hub, simplifying the deployment of AI models for vision, audio and speech applications to edge devices. This includes Llama3-TAIDE-LX-8B, one of several foundation models developed as a fundamental building block for Taiwan’s developer community.


TAIDE is redefining traditional Chinese language processing, and edge AI dialogue systems. Edge AI enables processing and inference to be performed locally, closer to where the data is generated, reducing the need for data transmission to the cloud. This results in faster response times, improved security, and reduced bandwidth requirements, as well as improved privacy, latency, reliability, flexibility and cost reductions.  

A Vision for Taiwan's Edge AI Leadership

TAIDE on Qualcomm AI Hub underscores Taiwan’s opportunity to lead in edge AI across technologies, applications and use cases available in Taiwan.

TAIDE on Qualcomm AI Hub aligns with Taiwan’s AI Plan 2.0 and will help to nurture local technology and industrial development. By offering a free, innovative tool to Taiwan’s developers, we are encouraging them to build more Taiwan-led Edge AI applications for companies in Taiwan and globally. Qualcomm Technologies is committed to enabling Taiwanese developers to create their own on-device end-to-end ML solutions with our partners across model makers, cloud providers, SDK providers, and others.

A Milestone for Taiwanese AI Innovation

The TAIDE foundation models were developed by the TAIDE team with support from the Taiwanese government. The initiative officially launched in April 2023 under the NSTC, leading to the release of the Llama3-TAIDE-LX-8B model in April 2024. Professor Tsai and his colleagues at National Central University (NCU) leveraged Qualcomm AI Hub to optimize the model and deploy it on Qualcomm platforms, including Snapdragon X Elite and Snapdragon 8 Elite.

The potential of Edge AI for Taiwan is vast as it enhances cost-effectiveness, sustainability and privacy by processing inputs and generating outputs directly on the device. As AI increasingly powers the user interface, the way in which we interact with our devices will change dramatically – resulting in more intuitive and personalized experiences, along with accelerated benefits for smaller users, including families, businesses and students.

Built on Proven Foundations

The Llama3-TAIDE-LX-8B-Chat-Alpha1 model is built on Meta’s Llama3 architecture and has been fine-tuned on with Traditional Chinese data. TAIDE delivers:

• Enhanced Office Task Capabilities: from drafting emails to summarizing documents, TAIDE streamlines everyday business communications.
  
  
• Robust Multi-turn Dialogue: engineered to handle complex conversations, ensuring meaningful and context-aware interactions.
  
  
• Advanced Translation Services: translate between Chinese and English, making cross-language communication more accessible than ever.

Verified Performance on Qualcomm AI Hub

Qualcomm AI Hub is dedicated to providing a trusted ecosystem where AI models are seamlessly deployed. TAIDE has been evaluated on the Qualcomm AI Hub platform, by running the model on real devices. Below are results running on the Snapdragon X Elite:

• Time to First Token: 0.21 to 6.77 seconds, resulting in swift response times for real-world applications
  
  
• Response Rate: delivering 10 tokens per second for fluid and dynamic interactions

Making TAIDE available on the Qualcomm AI Hub enables frequent testing of the model and acceleration of future application development to commercialization. These metrics not only validate TAIDE’s technical excellence but also empower developers to confidently integrate the model into their next-generation applications.

Join the Revolution

We invite developers to explore TAIDE on Qualcomm AI Hub and integrate it into their own applications. Take a look at our sample ChatApp for guidance on how to get started.

Join the Qualcomm AI Hub Slack channel to connect with the community, ask questions, and learn more.