MLC

Mlc is an open machine learning compiler framework designed to transform high-level ML models into highly optimized code for diverse hardware backends. Its primary purpose is to bridge the gap between popular ML frameworks and heterogeneous hardware, enabling efficient deployment on CPUs, GPUs, mobile chipsets, and edge devices. By automating low-level optimization, Mlc helps developers achieve high performance and portability without manually rewriting models for each platform.

Built on a modular compiler stack, Mlc supports graph-level and operator-level optimizations, automatic scheduling, and code generation targeting platforms such as CUDA, Vulkan, Metal, and various accelerators. It integrates with frameworks like PyTorch and TensorFlow via standardized model formats, allowing users to import existing models and compile them into optimized runtimes. Mlc also focuses on memory planning, kernel fusion, and quantization-aware compilation to reduce latency and resource consumption. Its design emphasizes reproducible performance tuning, enabling systematic exploration of optimization configurations.