online-qml 0.5.0

Online quantum machine learning package, using PyTorch.

online-qml

online-qml is a PyTorch based package for quantum extreme learning machines (QELM) simulations. It is with online shadow training readouts.

The distribution name is online-qml; the import name is online_qml.

Installation

Clone and sync with uv

Clone the repository, then sync the project environment:

git clone https://github.com/alex180500/online-qml.git
cd online-qml
uv sync

The default uv sync installs:

• package dependencies;
• development tools from the dev dependency group;
• the CPU PyTorch build from https://download.pytorch.org/whl/cpu.

To install the default CPU environment without development tools:

uv sync --no-dev

Supported PyTorch backends for local uv sync are:

Backend	Command
CPU	uv sync
CPU, no dev tools	uv sync --no-dev
CUDA 12.6	uv sync --no-group cpu --extra cu126
CUDA 12.8	uv sync --no-group cpu --extra cu128
CUDA 13.0	uv sync --no-group cpu --extra cu130

The cpu group is enabled by default. Disable it when selecting a CUDA extra, otherwise uv will reject the sync because two Torch backends were requested.

Install as a dependency with pip

When online-qml is installed with pip, only the package dependencies are installed. The local uv dependency groups, including dev and cpu, are not installed by downstream projects.

Install the desired PyTorch build first, then install online-qml:

python -m pip install --index-url https://download.pytorch.org/whl/cpu "torch>=2.11"
python -m pip install online-qml

For CUDA, use the matching PyTorch wheel index:

Backend	PyTorch install command
CPU	python -m pip install --index-url https://download.pytorch.org/whl/cpu "torch>=2.11"
CUDA 12.6	python -m pip install --index-url https://download.pytorch.org/whl/cu126 "torch>=2.11"
CUDA 12.8	python -m pip install --index-url https://download.pytorch.org/whl/cu128 "torch>=2.11"
CUDA 13.0	python -m pip install --index-url https://download.pytorch.org/whl/cu130 "torch>=2.11"

Then install online-qml:

python -m pip install online-qml

Do not rely on pip install "online-qml[cu130]" by itself to select the CUDA wheel index. Package extras can request torch, but pip package metadata cannot carry the uv source-index routing from this repository.

Current focus

• generate Haar pure states and random Naimark POVMs;
• train OST and prior-frame readout layers;
• train dense pseudoinverse/ridge baselines;
• evaluate Haar bias and variance;
• study state-frame and measurement-frame distances.

Minimal example

import torch
from online_qml.quantum import sample_dm, sample_povm, shots_outcome
from online_qml.estimators import ShadowReadoutEstimator

states = sample_dm(1000, d=2, dtype=torch.cdouble)
povm = sample_povm(16, d=2, dtype=torch.cdouble)
outcomes = shots_outcome(povm, states, shots=1)
obs = sample_dm(1, d=2, dtype=torch.cdouble).T

est = ShadowReadoutEstimator(
    n_out=16,
    d=2,
    dtype=torch.float64,
    methods=("ost", "state_prior_ost"),
)
est.update_single_shot(outcomes[:, 0], states)
layers = est.layers(obs)
W_ost = layers["ost"]
W_state_prior = layers["state_prior_ost"]