Catalyst is an experimental package designed for just-in-time (JIT) compilation of PennyLane programs, allowing for the compilation of entire quantum-classical workflows. It integrates directly with Python, enabling developers to decorate quantum code and hybrid functions with the @qjit decorator for significant performance improvements over standard Python execution. The software provides advanced control flow capabilities that support both quantum and classical instructions, along with infrastructure for both quantum and classical compilation. Catalyst is built to be end-to-end differentiable and supports high-performance simulators like PennyLane-Lightning and Amazon Braket devices, with future hardware support planned. The core components include the Catalyst Compiler, built with MLIR and a quantum dialect for optimized representation, and the Catalyst Runtime, a C++ runtime enabling execution on multiple devices via QIR. Additionally, a Python frontend for PennyLane and JAX offers the @qjit decorator and functions for defining Catalyst-compatible control flow, gradients, and mid-circuit measurements.

Features & Benefits

• End-to-End Compilation: Compiles the entire quantum-classical workflow, including optimization loops.
• Performance Improvement: Utilizes @qjit decorator for significant speed gains over standard Python.
• Advanced Control Flow: Supports complex control flow for both quantum and classical instructions.
• Differentiability: Designed for end-to-end auto-differentiation of hybrid programs.
• Device Support: Integrates with PennyLane-Lightning, Amazon Braket, and plans for additional hardware.