README.md

---
annotations_creators: []
language:
  - en
language_creators: []
license: []
multilinguality:
- monolingual
pretty_name: 'quantum_circuit_routing'
size_categories:
- 10K<n<100K
source_datasets:
- 'original'
tags:
  - adaption
  - instruction-tuning
  - code
  - science
task_categories: []
task_ids: []
---

![banner](https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/c68ac26d-586c-43f7-a1d2-edd7ce120ad5.png)

This dataset is a remastered version prepared using [Adaption's](https://adaptionlabs.ai/app/auth) Adaptive Data platform.

# quantum_circuit_routing

This dataset contains pairs of abstract OpenQASM 2.0 quantum circuits and their hardware-compliant compiled versions. Each sample specifies a target qubit topology (e.g., Star, Ring, Linear, Grid, HeavyHex) and coupling map constraints. The compiled outputs include necessary SWAP gates inserted to satisfy physical connectivity requirements while minimizing circuit depth.

> **Dataset Motto:**  
> *"Embedding physical spatial graph calculus into LLM weight space to eliminate graph hallucinations and enable deterministic quantum hardware circuit routing."*

### Dataset size

There are 15,000 data points in this dataset. This is an instruction tuning dataset.

### Quality of Remastered Dataset

The final quality is B, with a relative quality improvement of -26.0%.

### Domain
- Code (80%)
- Science (20%)

### Language
- English (100%)

### Tone
- Technical (100%)

### Evaluation Results

- **Quality Gains:**
<img src="https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/3e335eb7-a195-4de4-94f9-4b3f0a221690.png" alt="QualityGains" style="max-width: 50%; display: block; margin-left: auto; margin-right: auto;" />

- **Grade Improvement:**
<img src="https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/b9a712cd-8a96-47e0-b5a7-5b1d683479e9.png" alt="Grade" style="max-width: 50%; display: block; margin-left: auto; margin-right: auto;" />

- **Percentile Chart:**
<img src="https://proteus-prod-public.s3.us-east-1.amazonaws.com/temp/52761e13-0dea-4c29-a368-9cbac7fec0fe.png" alt="Percentile Chart" style="max-width: 50%; display: block; margin-left: auto; margin-right: auto;" />

---

### Usage & Quickstart

Load the dataset directly using the Hugging Face `datasets` library:

```python
from datasets import load_dataset

# Load the instruction-tuning dataset
dataset = load_dataset("jay2219/quantum-circuit-routing")

# Inspect a sample training pair
sample = dataset["train"][0]
print("System Role:\n", sample["messages"][0]["content"])
print("User Prompt (Abstract QASM & Topology):\n", sample["messages"][1]["content"])
print("Assistant Response (Routed Compliant QASM):\n", sample["messages"][2]["content"])
```

---

### Data Schema & Fields

Each record follows the standard OpenAI Chat format:
- **`messages[0]` (`system`)**: Enforces deterministic quantum hardware compilation role and strict OpenQASM output requirements.
- **`messages[1]` (`user`)**: Specifies the physical hardware topology, qubit count, explicit coupling map (valid edge list), hardware constraints, and input abstract OpenQASM 2.0.
- **`messages[2]` (`assistant`)**: Compiled OpenQASM 2.0 code containing minimal SWAP gate insertions to satisfy physical coupling map adjacency constraints.

---

### Hardware Topologies Covered

The 15,000 instruction-tuning samples span 5 distinct physical quantum hardware architectures:
1. **Linear:** 1D line chain topology (e.g. 5–15 qubits).
2. **Ring:** 1D closed-loop ring topology.
3. **Star:** Central hub routing graph (all peripheral qubits connect through hub node 0).
4. **Grid:** 2D rectangular lattice QPU connectivity graphs (e.g. 2x2 up to 5x3).
5. **HeavyHex:** IBM Heavy-Hex lattice architecture.

---

### Associated Models & Evaluation Benchmarks

- **Fine-Tuned Model:** [`jay2219/Q-Route-70B`](https://huggingface.co/jay2219/Q-Route-70B)
- **Held-Out Evaluation Benchmark:** [`jay2219/Q-Route-Benchmark`](https://huggingface.co/datasets/jay2219/Q-Route-Benchmark) (100 non-overlapping evaluation circuits with 0% data contamination)