PennyLane, the popular machine learning toolkit from Xanadu, has announced the latest beta release of Catalyst, the next-generation compilation framework. Catalyst allows the construction of hybrid quantum-classical programs that are automatically translated for easy processing on Quantum Processing Units, resulting in considerable speed and scalability gains and the possibility to explore new ways of programming quantum computers.
The Catalyst currently supports CPUs and the Lightning high-performance simulators; thus, customers can still deploy compiled processes to run alongside the hardware. And because of the compatibility of just-in-time (JIT) compilation and PennyLane, customers can still utilize the same PennyLane user interface, but this time, with the addition of experiencing significant speed improvements.
Incorporation of Just-in-Time (JIT) to the Python Language for sustainable GPUs.
There has been a drastic shift from the period of TensorFlow 1.0 to JAX, which seamlessly merges dynamic Python with under-the-hood just-in-time compilation.
Thus, PennyLane no longer needs to dust off C++ or Fortran memory for that much-needed speed bump; instead, they can decorate their existing Python code with @jit to gain access to performance, scalability, and even GPUs.
Catalyst gives an Advantage to Quantum Systems.
Catalyst’s ability to embed complicated control flow around quantum processes — such as if statements, for loops, and measurement feedforward is undoubtedly a huge advantage.
This enables the hardware to directly comprehend and apply control flow around quantum operations, resulting in a much faster compilation. Indeed, Catalyst is a true hybrid quantum-classical JIT that excels when used to build a complete process.
The Catalyst’s Structure
Catalyst presently consists of two components: the Catalyst Compiler and the Catalyst Runtime. The main Catalyst compiler is written in MLIR, with a quantum dialect added to express quantum instructions. This enables a high-level intermediate representation of the program’s classical and quantum components, resulting in advantages during optimization.
On the other hand, the runtime is a C++ QIR runtime that allows Catalyst-compiled quantum applications to be executed. The Catalyst documentation has a detailed list of the quantum instruction set supported by this runtime implementation.
PennyLane’s long-range plan for the ‘Catalyst.’
The PennyLane front will most likely be upstreamed into PennyLane proper as they continue to develop the Catalyst system, offering native JIT functionality incorporated into PennyLane. Furthermore, PennyLane has also looked into better ways to combine control flow with existing code, such as experimenting with native Python control syntax.
They have also planned to explore the possibilities of a compiler stack and adding quantum compilation methods. Lastly, PennyLane has also revealed that they plan to expand its compatibility for additional devices, including quantum hardware.
PennyLane continues to aspire to construct the quantum programming framework through the Catalyst; they are focused on a future in which the execution of arbitrary classical and quantum code alongside CPUs, GPUs, and QPUs works in tandem.
Read more about it here.