The MERA Compiler and Framework by EdgeCortix streamlines the deployment of neural network models across varied hardware architectures while maintaining efficiency and performance. MERA acts as a platform-agnostic toolset, featuring comprehensive APIs, code-generation capabilities, and runtime support, which facilitate the deployment of pre-trained deep neural networks. This compiler supports the integration of advanced AI applications in vision, audio, and language processing, helping developers optimize deployment workflows using familiar platforms.
MERA uniquely enhances integration ease across AMD, Intel, Arm, and RISC-V processors with built-in heterogeneous support, simplifying EdgeCortix AI platform assimilation into existing systems. Pre-defined models from Hugging Face or EdgeCortix Model Library are optimized through post-training calibration and quantization, making MERA an invaluable resource for AI inference development.
Beyond its software stack, MERA's contributions span comprehensive toolkits that include runtime configuration and simulation capabilities. This compiler empowers developers to scale AI inference from modeling to deployment effortlessly while achieving market-leading energy efficiencies when used alongside SAKURA-II modules.
