Embeddings for
Quantum Circuits
A lightweight transformer that converts quantum circuits into semantic vector representations, enabling search, deduplication, clustering, and fast equivalence filtering.
647K
Parameters
128-D
Embeddings
CPU
Fast
Open
Source
$ python similarity.py
from qiskit import QuantumCircuitfrom quark import CircuitEncoder, embed a = QuantumCircuit(3)a.h(0)a.cx(0,1) b = QuantumCircuit(3)b.h(0)b.h(0)b.h(0)b.cx(0,1) ea, eb = embed(model, [a, b])print((ea * eb).sum())# 0.96$ QxQuark Circuit Understanding Workflow
Transform quantum circuits into semantic vector representations for retrieval, clustering, and equivalence-aware analysis.
Quantum Circuits
Multi-Framework Ingestion
Qiskit · PennyLane · Cirq · OpenQASM
QxQuark Engine
Gate Tokenizer
Graph Neural Encoder
647K-Param Model
Vector Embeddings
Semantic Fingerprints
Similarity Metric
Nearest-Neighbor Index
Equivalence Check
Unitary Verification
CPU-Native · No GPU
PyTorch Backend
Python SDK & CLI
Deduplication
Group Equivalent Circuits
Semantic Search
Search by Example
Optimization QA
Transpiler Regression Checks
Clustering
Structural Circuit Discovery
Library Intelligence
Searchable Codebases
Key Capabilities
Semantic Search
Find circuits that compute similar operations even when implemented differently.
Circuit Deduplication
Identify duplicate circuits across repositories and benchmark datasets.
Equivalence Pre-Filtering
Filter candidates before expensive exact unitary verification.
Repository Discovery
Cluster and explore large collections of quantum circuits using embeddings.
Accelerate Quantum Circuit Analysis
0+
Circuits / Second
Generate embeddings efficiently on commodity CPU hardware.
0-D
Vector Representation
Compact semantic representation suitable for retrieval and indexing.
0
Verified Rewrite Families
Trained using equivalence-preserving circuit transformations.
The Problem
Which of these circuits are actually the same?
“QxQuark sits between fragile heuristics and expensive exact verification, providing fast semantic understanding of quantum circuits.”
How QxQuark Works
Tokenize
Each gate becomes a structured token.
Embed
Learned embeddings capture gate semantics.
Transformer Encoder
3 layers, 4 attention heads, CLS representation.
Normalize
Produces a 128-dimensional unit vector.
647K
Parameters
3
Transformer Layers
4
Attention Heads
128-D
Embeddings
CPU
Optimized
Benchmark Results
Real benchmark numbers with transparent reporting.
In-Distribution Recall@10
Held-Out Rewrite Recall@10
Gate vs Inverse Separation
QASMBench OOD Recall@10
Note — Ties the strongest baseline on out-of-distribution benchmarks while significantly outperforming on equivalence-aware retrieval tasks.
Research Highlights
Distinct Inverse Tokens
Separates S from S†, T from T†, and other inverse operations.
Hard Negative Training
Improves discrimination between near-equivalent circuits.
Expanded Rewrite Coverage
Supports additional verified equivalence-preserving transformations.
Industry & Research Applications
Interactive Playground
Two circuits in. One similarity score out.
Circuit A
Circuit B
Similarity Score
+1.000
Cosine similarity between the two 128-dimensional embeddings. Scores near +1.000 indicate semantically equivalent circuits; low or negative scores indicate the circuits compute different operations.
Open Source & Research
Build Smarter Quantum Circuit Workflows
Use semantic embeddings to search, organize, analyze, and understand quantum circuits at scale.