v0.1.0a17
Highlights
- Simplified the HTML report into a concise throughput and duration summary with visible confidence and expandable supporting evidence.
- Corrected documentation for scoring order, repeat aggregation, warmups, CLI workflows, local parallel validation, and supported Python versions.
- Refreshed example reports and screenshots to match the current report format.
Release artifacts
| Platform | Install |
|---|---|
| PyPI | pip install "vllm-optimizer==0.1.0a17" |
| Linux/WSL runtime | pip install "vllm-optimizer[runtime]==0.1.0a17" |
The universal wheel, source distribution, and SBOM are attached to the release.
Reports and documentation
- Report summaries now show throughput, duration, confidence, and compact settings first; complete trial data remains available in expandable details.
- Fixed stale guide links, simplified first-run setup, and distinguished published hardware evidence from pending validation and roadmap proposals.
- Updated the YAML configuration example and adaptive-repeat documentation.
Validation and limitations
The package version is 0.1.0a17, and the runtime reports the same version.
Documentation and package validation should be run for the release workflow.
Hardware smoke and real-backend validation remain dependent on the manual GPU
workflow.
Full changelog: v0.1.0a16...v0.1.0a17.