v0.3.0

AC-Prof v0.3.0

AC-Prof v0.3.0 is a substantial update focused on making containerized AI model profiling easier to use, more reproducible, and easier to audit. This release includes improvements beyond the merged PRs listed below, including changes made through direct commits.

Highlights

Easier setup and experiment workflows

• Unified CLI and simpler installation. Install AC-Prof from PyPI with uv tool install acprof==0.3.0 and start the terminal interface with acprof tui. Release assets also include a Linux x86_64 standalone executable. ([CLI](0a765e7), [packaging](865b232))
• Improved TUI experience. Added startup environment checks, clearer model preparation and experiment progress, planned-window information, recovery actions, and per-profiler controls. ([experiment workflows](ba3ec72), [environment checks](3742830), [recovery UI](b79b206))

More reproducible collection and stronger results

• More reliable runtime preparation. Reuse verified container images by default, apply explicit model-loading policies, and support controlled model download sources and byte budgets. ([image reuse](b30e0c7), [model policies](f272714), [download safeguards](247ca79))
• Auditable experiment outputs. Added versioned, layered CSV result projections, provenance and quality evidence, and commands to audit, migrate, and export results. Recovery paths now validate checkpoints before resuming post-processing. ([result layers](84c6d0f), [result audit](9c2a226), [recovery](92fdcb3))
• Richer measurements and comparisons. Expanded request-level latency, GPU/energy and CPU-counter measurement paths, deterministic resource-matrix planning, sampling diagnostics, and offline comparison, Pareto, and scaling reports. Availability of individual hardware counters depends on the host. ([measurement](3a9c5e8), [matrix and RAPL](fab4f25), [reports](073a20b))

Reliability, security, and release quality

• Strengthened input validation, container permissions, failed-run evidence, and recovery behavior.
• Moved development tooling to uv and expanded automated checks and distribution verification.
• Release publication requires verified native hardware evidence for the corresponding commit. ([hardware release gate](666e3fb))

Install and upgrade

uv tool install acprof==0.3.0
acprof --version
acprof tui

For existing installations, use uv tool upgrade acprof to upgrade to the latest published version. See the [installation guide](https://github.com/kainam15/AC-Prof/blob/v0.3.0/docs/getting_started/installation.md) for alternative installation methods and host setup.

Compatibility notes: AC-Prof requires Python 3.12+ when installed as a Python package. Full collection requires a supported native Linux host with a local Docker Engine and cgroup v2; hardware-specific profilers have additional requirements. WSL2 supports a more limited collection scope. The supported terminal interface entry point is acprof tui rather than the removed repository-specific TUI launcher.

Verification and downloads

Validation scope is recorded in the attached [verification.json](https://github.com/kainam15/AC-Prof/releases/download/v0.3.0/verification.json).

Hardware evidence: [verified GitHub Actions run](https://github.com/kainam15/AC-Prof/actions/runs/40167805238).

Release assets include the Python wheel, source archive, Linux x86_64 standalone executable, and SHA256SUMS. Check the download checksum before installing a downloaded release artifact.

Merged pull requests

• [#1](#1) — Simplify README quick start and remove redundant TUI instructions, by @kainam15.
• [#6](#6) — Migrate project tooling to uv, by @kainam15.
• [#7](#7) — Update the GitHub Actions dependency group, by @dependabot[bot].

New contributors: @kainam15 ([#1](#1)) and @dependabot[bot] ([#7](#7)).

Full changelog (including direct commits): [v0.2.0...v0.3.0](v0.2.0...v0.3.0).

v0.2.0

Full Changelog: v0.1.0...v0.2.0