Xanadu’s PennyLane, Qiskit, Cirq Benchmarked Against Divi Pipeline

The challenge of scaling quantum computing isn’t just building more qubits; it’s efficiently managing the thousands of nearly identical circuits required for a single variational step. Qoro has introduced divi, a new software development kit designed to address this bottleneck by treating batch generation as a composable pipeline, achieving up to 148 times faster pipeline execution against frameworks like PennyLane, Qiskit, and Cirq. “Useful work on NISQ hardware rarely means running one circuit; it means thousands of near-identical variants,” the company states, highlighting the scale of the problem, where N_groups × K × T × P circuits fan out from a single optimizer step. Divi maintains vendor agnosticism by outputting standard OpenQASM, offering both speed and flexibility for researchers seeking to optimize cloud compute costs and accelerate results.

Composable Circuit Transformations in Quantum Stacks

A single optimizer step in variational quantum algorithms can rapidly multiply the number of circuits needing execution, and this scale presents a significant hurdle for current quantum computing frameworks. While increasing qubit counts remains a central challenge, the bottleneck increasingly lies in efficiently preparing and submitting these massive circuit batches to near-term hardware. The problem isn’t simply building more qubits, but managing the exponential growth of circuit variants required for practical applications. Several independent efforts within the quantum software ecosystem have begun to address this challenge, each offering partial solutions. Xanadu’s PennyLane introduced composable circuit transformations, while IBM’s Qiskit undertook a substantial internal engineering effort, rewriting its circuit core in Rust (qiskit. _accelerate) to expose a primitive model at a lower level. Google’s Cirq implemented parameter sweeps as a first-class execution primitive, and Mitiq developed error mitigation techniques designed to be frontend-agnostic.

OpenQASM 3 is standardizing parametric programs through input declarations. However, these advancements remain fragmented, and integrating them into a cohesive, scalable pipeline has proven difficult. Qoro is now presenting divi, a software development kit designed to compose these existing techniques into a unified circuit-batch generation pipeline. The core principle is to treat the entire batch as a single, composable pipeline operating on a templated circuit. The company states that this approach defers parameter binding until the final serialization step, allowing structural elements like grouping, folding, and twirling to be applied to a parameter-free template. Benchmarking against PennyLane, Qiskit, and Cirq, Qoro claims divi achieves up to 148 times faster pipeline execution by amortizing the cost of structural transformations across the entire batch. “The expensive, parameter-independent work happens once and is shared across the whole fan-out,” they assert.

diviSDK Pipeline for Accelerated Batch Generation

The pursuit of scalable quantum computation has increasingly focused attention on the efficiency of preparing and submitting large batches of circuits, rather than simply increasing qubit counts. Current quantum frameworks, while capable of generating individual circuits, often struggle with the exponential growth in computational demands when scaling to the thousands of near-identical variants required for variational algorithms like VQE and QAOA. This inefficiency stems from repeatedly rebuilding and serializing each circuit in a batch, which can lead to increased cloud GPU costs. Divi’s core innovation lies in its ability to defer parameter binding until the final stage of the process. This contrasts with many existing frameworks where parameter sweeps are materialized into multiple copies of the circuit long before submission. The diviSDK pipeline composes an ordered sequence of stages, grouping, folding, twirling, and basis rotation, operating on a parameter-free template.

The circuit body is serialized only once, with parameter sets then substituted efficiently. This strategy ensures that the computationally expensive, parameter-independent work is performed only once and shared across all circuit variants. The system measures performance at two key points: in-memory batch generation and fully serialized batch output, with benchmarks against PennyLane, Qiskit, and Cirq. Qoro reports that divi achieves up to 148 times faster pipeline execution by adopting this composable approach. “Where other frameworks lock your IP into proprietary serialization schemes, like QPY for Qiskit or custom protobufs for Google, Qoro outputs standard OpenQASM,” the company states. Qiskit’s rewriting its circuit core in Rust (qiskit. _accelerate), and Cirq’s parameter sweep execution all informed the development of divi. However, divi uniquely integrates these concepts into a unified pipeline optimized for batch generation.

The result, according to Qoro, is faster generation, reduced cloud compute costs, accelerated time-to-result, and greater flexibility in selecting hardware providers. Divi is available as an open-source package, and Qoro offers a commercial platform, Qoro Solo, for users to test the pipeline with real-scale quantum algorithms.

Benchmarking divi Against PennyLane, Qiskit, and Cirq

Researchers at Qoro are tackling a critical bottleneck in near-term quantum computing: the efficient generation of large circuit batches required for variational algorithms. The core of divi’s approach lies in deferring parameter binding until the final serialization stage. Divi achieves this through a system that leverages composable transforms, broadcast parameter handling, and late binding, resulting in up to a 148 times speedup in pipeline execution. Importantly, Qoro emphasizes vendor agnosticism, outputting standard OpenQASM, avoiding the proprietary serialization schemes found in other systems. The benchmark focused on in-memory generation, the time taken to construct the complete batch in each framework’s native representation, and then measured fully serialized batches. The results reveal significant performance differences. PennyLane’s transform stack re-runs on every execution, creating a substantial generation gap, while Qiskit and Cirq exhibit more modest, though still noticeable, delays.

The scale of the improvement is particularly evident when considering the number of circuits generated; a single optimizer step fans out into N_groups × K × T × P circuits. Divi’s ability to defer structural calculations to a parameter-free template, rather than rebuilding per circuit, accounts for much of the performance gain. The team highlights that the ecosystem already possesses the necessary components for optimization; “Each framework supplies a brilliant piece—composable transforms, primitive broadcasting, sweep execution, frontend-agnostic mitigation—but assembling them into the batch is left to the user.” Divi’s open-source nature, accessible via pip install qoro-divi, offers researchers a tool to accelerate their variational quantum algorithms and reduce cloud compute costs.

OpenQASM 3 and Parametric Program Standardization

The increasing complexity of quantum algorithms demands a shift in how circuits are generated and managed, moving beyond simply increasing qubit counts to address the challenges of scaling variational quantum algorithms. A key component of this evolution lies in the standardization of parametric programs. While frameworks like Xanadu’s PennyLane have introduced and IBM’s Qiskit has rewritten its circuit core in Rust (qiskit. _accelerate) to expose a primitive model, these are often isolated improvements. Divi’s approach, however, aims to compose these pieces into a unified pipeline, deferring parameter binding to minimize redundant calculations. This approach ensures that the computationally expensive, parameter-independent work is performed only once and shared across all circuit variants.

In-Memory Batch Generation Performance Comparison

The assumption that scaling quantum computing hinges solely on increasing qubit counts overlooks a critical, often more pressing, challenge: efficiently preparing the vast number of circuits required for even a single step in complex algorithms. While hardware advances continue, the software infrastructure for managing these “batches” of circuits has lagged, creating a significant bottleneck. However, these improvements are often isolated. Qoro, a company focused on quantum software, has addressed this fragmentation with divi, a new SDK designed to compose these individual pieces into a unified, high-performance pipeline. This approach means that the computationally expensive, parameter-independent work is performed only once and shared across all circuit variants. The performance gains are particularly noticeable when considering the scale of these batches. Divi’s architecture tackles this exponential growth by operating on a parameter-free template, serializing the circuit structure once, and then efficiently substituting parameter sets. Qoro outputs standard OpenQASM, offering users the flexibility to route workloads to the most cost-effective and available quantum processing unit.