AMD Ryzen AI Software 1.5: More Models, Faster Deployments, and Better Experience

Introduction

The Ryzen™ AI 1.5 Software release is here - and it’s packed with updates that make AI development faster, more flexible, and more efficient. This includes expanded support for LLMs, Stable Diffusion, and newly released ASR pipelines. Additionally, we’ve got faster compile times and a smaller SDK footprint. 

New Support for Stable Diffusion and ASR Pipelines

We’ve added optimized support for SD 1.5, SD 2.1, SDXL, and SD 3.0 pipelines. This gives developers more flexibility to choose the right model for their creative or generative AI use case—whether it’s image generation, inpainting, or fine-tuned visual tasks. We’ve also added new preview support for Zipformer, Whisper-base, and Whisper-tiny. These models will make it easier for developers to build speech-to-text applications with lightweight, high-performance models. These models are ideal for real-time transcription, voice interfaces, and accessibility features such as live captions. This is available for Preview in our Early Access Site.

Preemption: Smarter Resource Management

Preemption is now available to pause or interrupt lower priority tasks to free resources for higher priority ones. This feature is useful in multi-model environments or on resource-constrained devices, where responsiveness is critical. With preemption, developers can prioritize latency-sensitive tasks like real-time inference or user-facing interactions without waiting for background jobs to complete. System responsiveness can be improved by dynamically reallocating compute resources to the most critical workloads.  

ONNX Context Caching = Faster Deployments

ONNX context caching eliminates the need to recompile models every time they run. Without context caching, deploying an ONNX model involves compiling or optimizing the model graph before execution - a process that can be time-consuming, especially for large or complex models. With context caching, the results are saved and reused, allowing subsequent runs to skip the setup and launch immediately. This reduces startup time, conserves compute and memory resources, and speeds up deployment, especially valuable in production or edge environments.

Developer Experience Boosts

Not only has the Ryzen AI Software package reduced in size, meaning smaller downloads and reduced file counts, but compile times have also improved with 2× faster for CNNs and 3× faster for Transformers.

Summary

We are excited to announce continuous improvements to Ryzen AI software and encourage developers to provide feedback on how these updates enhance their experience. For a detailed overview of the new features and enhancements in the 1.5 software release, check out the official release notes. Subscribe to be notified of future Ryzen AI software updates and get the latest tools and resources to help you explore the limits of what's possible on AI PCs."

Additional Resources

• Ryzen AI Video Tutorials 
• AMD Ryzen AI Developer Hub
• Ryzen AI 1.5 Release Notes
  
• Ryzen AI 1.5 Models
  • Hybrid: RyzenAI-1.5_LLM_Hybrid_Models - a amd Collection
  • NPU-Only: RyzenAI-1.5_LLM_NPU_Models - a amd Collection