Skip to main content

AI Inference Software Stack

From model to silicon, one unified stack

ZMC

Model Compiler

Based on MLIR framework, seamlessly integrated with LLVM
Support for importing model files from mainstream AI frameworks including PyTorch, ONNX, JAX, and TensorFlow
Support Triton kernels from multiple source, including FlagGems, PyTorch, and user-defined
Unified optimization for computation graph and kernel fusion
Support both RISC-V CPUs and heterogeneous compute accelerators

ZTC

Triton Compiler

Based on MLIR framework, seamlessly integrated with LLVM
Support for AI kernels written in Triton while achieving comparable performance to those written in C
Support both RISC-V CPUs and heterogeneous compute accelerators