mirage-project/mirage
A compiler and runtime system that fuses multi-GPU LLM inference computation and communication into a single megakernel for lower latency.
Not currently ranked — collecting fresh signals.
star history
Mirage Persistent Kernel (MPK) is a compiler and runtime that automatically transforms LLM inference workflows from multiple separate kernels into a single fused megakernel. It merges computation and communication across multiple GPUs into one kernel launch, achieving 1.2× to 6.7× latency reduction. Developers provide a few dozen lines of Python to define kernel inputs and outputs, and the system compiles models from Hugging Face into optimized megakernels using Triton and FlashInfer backends.
Frequently asked
- What is mirage-project/mirage?
- A compiler and runtime system that fuses multi-GPU LLM inference computation and communication into a single megakernel for lower latency.
- Is mirage open source?
- Yes — mirage-project/mirage is open source, released under the Apache-2.0 license.
- What language is mirage written in?
- mirage-project/mirage is primarily written in Cuda.
- How popular is mirage?
- mirage-project/mirage has 2.4k stars on GitHub.
- Where can I find mirage?
- mirage-project/mirage is on GitHub at https://github.com/mirage-project/mirage.