microsoft/qdk
Publicmirrored fromhttps://github.com/microsoft/qdkAvailable
samples/python_interop/qiskit_submission_to_azure.ipynb
280lines · modecode
| 1 | { |
| 2 | "cells": [ |
| 3 | { |
| 4 | "cell_type": "markdown", |
| 5 | "id": "90f546a9", |
| 6 | "metadata": {}, |
| 7 | "source": [ |
| 8 | "# Submitting Qiskit Circuits to Azure Quantum with the QDK\n", |
| 9 | "\n", |
| 10 | "This notebook demonstrates how to run Qiskit `QuantumCircuit` jobs on Azure Quantum using `AzureQuantumProvider` from the QDK. It also shows how to handle parameterized circuits by binding parameters prior to execution." |
| 11 | ] |
| 12 | }, |
| 13 | { |
| 14 | "cell_type": "markdown", |
| 15 | "id": "2b838c23", |
| 16 | "metadata": {}, |
| 17 | "source": [ |
| 18 | "The workflow demonstrated here:\n", |
| 19 | "\n", |
| 20 | "1. Build a Qiskit `QuantumCircuit`.\n", |
| 21 | "2. Reference an existing Azure Quantum workspace with `qdk.azure.Workspace`.\n", |
| 22 | "3. Instantiate `AzureQuantumProvider(workspace)` and browse available backends (`provider.backends()`).\n", |
| 23 | "4. Pick a backend and call `backend.run(circuit, shots=...)` and fetch results (`job.result().get_counts()`)." |
| 24 | ] |
| 25 | }, |
| 26 | { |
| 27 | "cell_type": "markdown", |
| 28 | "id": "004e5982", |
| 29 | "metadata": {}, |
| 30 | "source": [ |
| 31 | "## Prerequisites\n", |
| 32 | "\n", |
| 33 | "This notebook assumes the `qdk`, Azure Quantum integration, and Qiskit packages are installed. You can install everything (including the Azure + Qiskit extras) with:" |
| 34 | ] |
| 35 | }, |
| 36 | { |
| 37 | "cell_type": "code", |
| 38 | "execution_count": null, |
| 39 | "id": "dfa160e8", |
| 40 | "metadata": {}, |
| 41 | "outputs": [], |
| 42 | "source": [ |
| 43 | "%pip install \"qdk[azure,qiskit]\"" |
| 44 | ] |
| 45 | }, |
| 46 | { |
| 47 | "cell_type": "markdown", |
| 48 | "id": "b7f37bcd", |
| 49 | "metadata": {}, |
| 50 | "source": [ |
| 51 | "This installs:\n", |
| 52 | "- The base qdk package (compiler, OpenQASM/QIR tooling)\n", |
| 53 | "- Azure Quantum client dependencies for submission\n", |
| 54 | "- Qiskit for circuit construction\n", |
| 55 | "\n", |
| 56 | "After installing, restart the notebook kernel if it was already running. You can verify installation with:" |
| 57 | ] |
| 58 | }, |
| 59 | { |
| 60 | "cell_type": "code", |
| 61 | "execution_count": null, |
| 62 | "id": "f7446ed1", |
| 63 | "metadata": {}, |
| 64 | "outputs": [], |
| 65 | "source": [ |
| 66 | "import qiskit, qdk, qdk.azure # should import without errors" |
| 67 | ] |
| 68 | }, |
| 69 | { |
| 70 | "cell_type": "markdown", |
| 71 | "id": "4f761649", |
| 72 | "metadata": {}, |
| 73 | "source": [ |
| 74 | "## Build a simple Qiskit circuit\n", |
| 75 | "\n", |
| 76 | "We start with a Bell-state circuit — a Hadamard on qubit 0 followed by a CNOT — which produces the entangled state $\\frac{1}{\\sqrt{2}}(|00\\rangle + |11\\rangle)$. After constructing the circuit we'll submit it to an Azure Quantum target." |
| 77 | ] |
| 78 | }, |
| 79 | { |
| 80 | "cell_type": "code", |
| 81 | "execution_count": null, |
| 82 | "id": "e91dd2d7", |
| 83 | "metadata": {}, |
| 84 | "outputs": [], |
| 85 | "source": [ |
| 86 | "from qiskit import QuantumCircuit\n", |
| 87 | "\n", |
| 88 | "q = QuantumCircuit(2, 2)\n", |
| 89 | "q.h(0)\n", |
| 90 | "q.cx(0, 1)\n", |
| 91 | "q.measure([0, 1], [0, 1])\n", |
| 92 | "\n", |
| 93 | "circuit = q\n", |
| 94 | "circuit.draw(output=\"text\")" |
| 95 | ] |
| 96 | }, |
| 97 | { |
| 98 | "cell_type": "markdown", |
| 99 | "id": "efd50528", |
| 100 | "metadata": {}, |
| 101 | "source": [ |
| 102 | "## Configure Azure Quantum workspace connection\n", |
| 103 | "\n", |
| 104 | "To connect to an Azure workspace replace the following variables with your own values." |
| 105 | ] |
| 106 | }, |
| 107 | { |
| 108 | "cell_type": "code", |
| 109 | "execution_count": null, |
| 110 | "id": "54206752", |
| 111 | "metadata": {}, |
| 112 | "outputs": [], |
| 113 | "source": [ |
| 114 | "subscription_id = 'xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx'\n", |
| 115 | "resource_group = 'myresourcegroup'\n", |
| 116 | "workspace_name = 'myworkspace'\n", |
| 117 | "location = 'westus'" |
| 118 | ] |
| 119 | }, |
| 120 | { |
| 121 | "cell_type": "markdown", |
| 122 | "id": "cdae59f7", |
| 123 | "metadata": {}, |
| 124 | "source": [ |
| 125 | "## Submit the circuit to Azure Quantum\n", |
| 126 | "\n", |
| 127 | "`AzureQuantumProvider` exposes Azure Quantum targets as Qiskit backends. When you call `provider.get_backend(target_name)`, the SDK returns one of two kinds of backend:\n", |
| 128 | "\n", |
| 129 | "- **Provider-specific backends** — Some hardware vendors (e.g. IonQ, Quantinuum) ship dedicated Qiskit backend classes that have native integration with their APIs. These handle gate translation and result parsing using hardware-specific logic.\n", |
| 130 | "- **Generic QIR backends** — For any other target that accepts QIR input, the SDK automatically wraps it as a generic backend. These compile the Qiskit circuit to OpenQASM 3 and then to QIR internally, so you don't have to manage those steps manually.\n", |
| 131 | "\n", |
| 132 | "In practice `get_backend()` selects the right type for you — you use the same call regardless of which backend type is returned." |
| 133 | ] |
| 134 | }, |
| 135 | { |
| 136 | "cell_type": "code", |
| 137 | "execution_count": null, |
| 138 | "id": "d9a0f59b", |
| 139 | "metadata": {}, |
| 140 | "outputs": [], |
| 141 | "source": [ |
| 142 | "from qdk.azure import Workspace\n", |
| 143 | "from qdk.azure.qiskit import AzureQuantumProvider\n", |
| 144 | "\n", |
| 145 | "workspace = Workspace(\n", |
| 146 | " subscription_id=subscription_id,\n", |
| 147 | " resource_group=resource_group,\n", |
| 148 | " name=workspace_name,\n", |
| 149 | " location=location,\n", |
| 150 | ")\n", |
| 151 | "\n", |
| 152 | "provider = AzureQuantumProvider(workspace)\n", |
| 153 | "\n", |
| 154 | "# List available backends\n", |
| 155 | "for backend in provider.backends():\n", |
| 156 | " print(f\"{backend.name:45s} {type(backend).__name__}\")" |
| 157 | ] |
| 158 | }, |
| 159 | { |
| 160 | "cell_type": "code", |
| 161 | "execution_count": null, |
| 162 | "id": "ce1b1809", |
| 163 | "metadata": {}, |
| 164 | "outputs": [], |
| 165 | "source": [ |
| 166 | "# Replace with any backend name from the list above\n", |
| 167 | "target_name = \"rigetti.sim.qvm\"\n", |
| 168 | "backend = provider.get_backend(target_name)\n", |
| 169 | "\n", |
| 170 | "job = backend.run(circuit, shots=100, job_name=\"qiskit-provider-job\")\n", |
| 171 | "print(f\"Job {job.job_id()} submitted — waiting for results...\")\n", |
| 172 | "\n", |
| 173 | "counts = job.result().get_counts()\n", |
| 174 | "total = sum(counts.values())\n", |
| 175 | "print(f\"\\nResults ({total} shots):\")\n", |
| 176 | "for bitstring, count in sorted(counts.items()):\n", |
| 177 | " print(f\" {bitstring}: {count:4d} ({count/total:.1%})\")" |
| 178 | ] |
| 179 | }, |
| 180 | { |
| 181 | "cell_type": "markdown", |
| 182 | "id": "9c84df75", |
| 183 | "metadata": {}, |
| 184 | "source": [ |
| 185 | "## Parameterized Qiskit circuits\n", |
| 186 | "\n", |
| 187 | "Many algorithms use symbolic parameters. Before submitting to azure (or before compiling to QIR), all circuit parameters must be bound to numeric values. Below we build a simple entangling ladder, apply a rotation parameterized by `θ` across all qubits, then uncompute the entanglement and measure the first qubit." |
| 188 | ] |
| 189 | }, |
| 190 | { |
| 191 | "cell_type": "code", |
| 192 | "execution_count": null, |
| 193 | "id": "5d766661", |
| 194 | "metadata": {}, |
| 195 | "outputs": [], |
| 196 | "source": [ |
| 197 | "from qiskit import QuantumCircuit\n", |
| 198 | "from qiskit.circuit import Parameter\n", |
| 199 | "\n", |
| 200 | "def get_parameterized_circuit(n = 5) -> QuantumCircuit:\n", |
| 201 | " theta = Parameter(\"θ\")\n", |
| 202 | " qc = QuantumCircuit(n, 1)\n", |
| 203 | " qc.h(0)\n", |
| 204 | " for i in range(n - 1):\n", |
| 205 | " qc.cx(i, i + 1)\n", |
| 206 | " qc.rz(theta, range(n))\n", |
| 207 | "\n", |
| 208 | " for i in reversed(range(n - 1)):\n", |
| 209 | " qc.cx(i, i + 1)\n", |
| 210 | " qc.h(0)\n", |
| 211 | " qc.measure(0, 0)\n", |
| 212 | " return qc\n", |
| 213 | "\n", |
| 214 | "# Build the symbolic (parameterized) circuit\n", |
| 215 | "parameterized_circuit = get_parameterized_circuit()\n", |
| 216 | "\n", |
| 217 | "# Bind θ to a numeric value, then visualize the bound circuit\n", |
| 218 | "bound_circuit = parameterized_circuit.assign_parameters({\"θ\": 0.5})\n", |
| 219 | "\n", |
| 220 | "bound_circuit.draw(output=\"text\")" |
| 221 | ] |
| 222 | }, |
| 223 | { |
| 224 | "cell_type": "markdown", |
| 225 | "id": "ee4d57b4", |
| 226 | "metadata": {}, |
| 227 | "source": [ |
| 228 | "## Submitting a bound parameterized circuit\n", |
| 229 | "\n", |
| 230 | "Here we submit the `bound_circuit` from above using the recommended `AzureQuantumProvider` approach. The printed result shows the measurement counts. Change the value of `θ` (or loop over several values) to explore how the outcome distribution varies." |
| 231 | ] |
| 232 | }, |
| 233 | { |
| 234 | "cell_type": "code", |
| 235 | "execution_count": null, |
| 236 | "id": "d9bb3eb0", |
| 237 | "metadata": {}, |
| 238 | "outputs": [], |
| 239 | "source": [ |
| 240 | "parameterized_job = backend.run(bound_circuit, shots=100, job_name=\"qiskit-parameterized-job\")\n", |
| 241 | "print(f\"Job {parameterized_job.job_id()} submitted — waiting for results...\")\n", |
| 242 | "\n", |
| 243 | "parameterized_counts = parameterized_job.result().get_counts()\n", |
| 244 | "print(parameterized_counts)" |
| 245 | ] |
| 246 | }, |
| 247 | { |
| 248 | "cell_type": "markdown", |
| 249 | "id": "0ce5d167", |
| 250 | "metadata": {}, |
| 251 | "source": [ |
| 252 | "## Handling qubit loss on noisy hardware\n", |
| 253 | "\n", |
| 254 | "On some hardware backends — particularly neutral-atom and trapped-ion devices — a qubit may be lost before measurement (e.g. an atom is ejected from the trap). When this happens, the backend records `\"-\"` in the bitstring position for that qubit rather than `\"0\"` or `\"1\"`. Because loss shots contain non-binary characters, they cannot be included in standard counts or probability distributions, which assume a fixed binary alphabet. The SDK therefore separates them automatically.\n", |
| 255 | "\n", |
| 256 | "The result data (`job.result().results[0].data`) exposes two parallel sets of fields:\n", |
| 257 | "\n", |
| 258 | "| Field | What it contains |\n", |
| 259 | "|---|---|\n", |
| 260 | "| **`counts`** | Bitstring → shot count, accepted shots only (no `\"-\"`) |\n", |
| 261 | "| **`probabilities`** | Bitstring → empirical probability, accepted shots only |\n", |
| 262 | "| **`memory`** | Per-shot bitstring list, accepted shots only |\n", |
| 263 | "| **`raw_counts`** | Bitstring → shot count, all shots (loss shots have `\"-\"`) |\n", |
| 264 | "| **`raw_probabilities`** | Bitstring → empirical probability, all shots |\n", |
| 265 | "| **`raw_memory`** | Per-shot bitstring list, all shots |\n", |
| 266 | "\n", |
| 267 | "Use the **accepted** fields (`counts`, `probabilities`, `memory`) for any downstream analysis that expects clean binary bitstrings. Use the **raw** fields (`raw_counts`, `raw_probabilities`, `raw_memory`) to inspect loss patterns — for example, to calculate the overall loss rate or identify which qubit positions are being lost most frequently.\n", |
| 268 | "\n", |
| 269 | "> **Tip**: A high loss rate may indicate hardware instability or a circuit that is too deep for the current calibration." |
| 270 | ] |
| 271 | } |
| 272 | ], |
| 273 | "metadata": { |
| 274 | "language_info": { |
| 275 | "name": "python" |
| 276 | } |
| 277 | }, |
| 278 | "nbformat": 4, |
| 279 | "nbformat_minor": 5 |
| 280 | } |
| 281 | |