| Documentation | Users Forum | #sig-spyre |
IBM Spyre is the first production-grade Artificial Intelligence Unit (AIU) accelerator born out of the IBM Research AIU family, and is part of a long-term strategy of developing novel architectures and full-stack technology solutions for the emerging space of generative AI. Spyre builds on the foundation of IBM's internal AIU research and delivers a scalable, efficient architecture for accelerating AI in enterprise environments.
spyre-inference is a vLLM platform plugin that enables seamless integration of IBM Spyre accelerators with vLLM via the torch-spyre PyTorch backend. It is the next evolution of sendnn-inference, leveraging PyTorch's native Inductor compiler backend through vLLM's plugin architecture.
For more information, check out the following:
- 📚 Meet the IBM Artificial Intelligence Unit
- 📽️ AI Accelerators: Transforming Scalability & Model Efficiency
- 🚀 Spyre Accelerator for IBM Z
- 🚀 Spyre Accelerator for IBM POWER
Visit our documentation:
We welcome and value any contributions and collaborations. Please check out Contributing to Spyre Inference for how to get involved.
You can reach out for discussion or support in the #sig-spyre channel in the vLLM Slack workspace or by opening an issue.
