Installation#

For users#

Installation#

The package works with Python 3.11+ and can be installed from both PyPI and/or conda-forge.

To install the package using the pip package manager, run the following command:

$ pip install bluesky-adaptive

To install the package using the conda package manager, run the following command:

$ conda install -c conda-forge bluesky-adaptive

Optional Extras#

bluesky-adaptive is modular — install only what you need by appending one or more extras:

Extra	Installs	Notes
`bluesky-adaptive[ax]`	`ax-platform`, `botorch`, `gpytorch`, `torch`	GPU torch by default; pair with `[cpu]` for CPU-only
`bluesky-adaptive[xopt]`	`xopt`, `botorch`, `gpytorch`, `torch`	GPU torch by default; pair with `[cpu]` for CPU-only
`bluesky-adaptive[queueserver]`	`bluesky-queueserver-api`	Transport layer only; pair with `[ax]` for `QueueserverAgent`
`bluesky-adaptive[cpu]`	(uv index routing)	Routes `torch` to the CPU-only PyTorch index; requires `uv`
`bluesky-adaptive[all]`	All backends + `[queueserver]`	No dev tooling; will grow as new backends are added
`bluesky-adaptive[dev]`	`bluesky-adaptive[all]` + dev tooling	For contributors

PyTorch Acceleration Options#

By default, bluesky-adaptive installs PyTorch with GPU support (~7GB). For environments without GPU support, or to reduce installation size, you can install a CPU-only version (~900MB) using uv:

$ pip install uv
$ uv pip install bluesky-adaptive[cpu]

This is particularly useful for:

Containerized deployments without GPU access
CI/CD pipelines
Development environments on laptops without NVIDIA GPUs
Edge computing scenarios

Note

The CPU-only installation requires uv, a fast Python package installer. If you prefer to use standard pip, the default installation will include GPU support.

For conda users who want CPU-only PyTorch:

$ conda install -c conda-forge bluesky-adaptive pytorch cpuonly -c pytorch