Meet Iris — a lightweight open-source library for multi-GPU programming in Triton.
Iris is about 370 lines of Python + Triton and exposes SHMEM-like Remote Memory Access (RMA) primitives — load, store, and atomics — for GPUs. The repository includes examples from simple memory operations to distributed GEMM with fused compute and concurrency patterns, so you can get started quickly.
The project is fully open source. More examples are on the way, and contributions are welcome — issues, discussions, and PRs all help shape where Iris goes next.
mawad•4mo ago
Iris is about 370 lines of Python + Triton and exposes SHMEM-like Remote Memory Access (RMA) primitives — load, store, and atomics — for GPUs. The repository includes examples from simple memory operations to distributed GEMM with fused compute and concurrency patterns, so you can get started quickly.
The project is fully open source. More examples are on the way, and contributions are welcome — issues, discussions, and PRs all help shape where Iris goes next.