v0.5.0

Release with a pip package

What's Changed

New UQ methods added:

• CoCoA by @rvashurin in #295
• Focus by @alfekka in #304
• KLE by @SpeedOfMagic in #278
• EigenScore by @ArtemVazh in #305
• LUQ by @ArtemVazh in #302
• CSL and RAUQ by @ArtemVazh in #347
• SelfCertainty by @cant-access-rediska0123 in #350

Other changes:

• Support for Python 3.11 by @SpeedOfMagic in #259
• vLLM inference support by @ArtemVazh in #308
• Visual LLM support by @alfekka in #341
• Calibration metrics (ECE and PCC) added to UEMetrics by @rvashurin in #386
• Support for greybox models from OpenAI API by @yobeen in #303
• Any statistic can be added to saved manager during benchmark run using save_stats parameter, added by @silvimica in #371

Internal changes:

• Stat Calculators are now built by factories and share resources, added by @IINemo in #249
• Better logging by @IINemo in #266

Full Changelog: https://github.com/IINemo/lm-polygraph/compare/v0.4.0..v0.5.0