abrax 1.1.0

Cross platform circuit transpiler

abraxas

A tiny library to transpile quantum circuits. The goal is to speed up the time it takes to convert simple circuits across languages fast. Anything beyond a certain complexity should be written in the respective languages directly.

Qiskit • CudaQ • Pennylane • Cirq • TKet • Quil • Braket

Install

While not explicity marked as required, it will be convenient to have qiskit installed since some functions use it to generate qasm code.

pip install abrax

Examples

toPennylane

from qiskit import QuantumCircuit, Parameter
from abrax import toPenny, toQasm
import pennylane as qml

qc = QuantumCircuit(3)
p = Parameter('x')
qc.h([0, 1, 2])
qc.cx(0, 2)
qc.rx(0, 0)
qc.cx(1, 2)
qc.ry(p, 2)

qasm = toQasm(qc)
dev = qml.device("default.qubit", wires=2)
circuit = toPenny(qasm, dev) # pennylane needs 'dev'
# print(qml.draw(circuit)())
# 0: ──H─╭X──RX(0.00)───────────────┤  Probs
# 1: ──H─│────────────╭X────────────┤  Probs
# 2: ──H─╰●──X────────╰●──RY(x)─────┤  Probs

Supported conversions:

graph LR
  subgraph Circuit
    direction LR
    IR[Pseudo-QASM]
    IR2[QASM]
  end
  A0[Qiskit] --> Circuit
  B0[Pennylane] --> Circuit
  C0[Cirq] --> Circuit
  D0[TKet] --> Circuit
  E0[Quil] --> Circuit
  F0["CudaQ (half)"] --> Circuit
  G0[Braket] --> Circuit

  Circuit --> A1[Qiskit]
  Circuit --> B1[Pennylane]
  Circuit --> C1[Cirq]
  Circuit --> D1[TKet]
  Circuit --> E1[Quil]
  Circuit --> F1[CudaQ]
  Circuit --> G1[Braket]

Known Issues

• Don't parameterise 'exponent' in cirq for variational circuits. Else it will hard code in some random value. As a slow hack you can run the transpiler in a for loop everytime the exponent/rads change.
• Compiling "TO" pyquil isn't supported yet, I don't want to write a full generator loop
• CudaQ cannot do variational gates "toQasm" and writing a parser for QIR/MLIR is extremely expensive

Supported Frameworks