v0.4.1

What's Changed

• fix: keep ROUGE and BLEU scores when a prediction is empty by @raashish1601 in #259
• fix: handle empty results in retrieval precision by @Thebearcoding in #257
• Fix invalid FAISS neighbors in multimodal retrieval by @Mrkinggggg in #256
• Fix getattr to catch KeyError in addition to AttributeError by @Jah-yee in #255
• fix: resolve BART and T5 refiner models in get_refiner by @kabishou11 in #254
• fix: raise AttributeError for missing Item and Dataset attributes by @kabishou11 in #253
• fix: score unparsable GAOKAO-MM multiple-choice answers as zero by @kabishou11 in #252
• fix: avoid matching empty references to every prediction by @jackiectl2 in #271
• Fix answer resolution for mixed choice and free-answer datasets by @March-7 in #270
• Preserve multimodal prompts when exporting dataset items by @Darssal in #269
• Fix cache reuse for Chinese ROUGE scoring by @Ganesh1729-ui in #261
• fix: use a consistent cache key in ZH_Rouge_Score.calculate_rouge by @Mrkinggggg in #262
• fix: parse GAOKAO-MM answers whose output starts with the answer marker by @kabishou11 in #266
• Fix loading saved JSON dataset arrays by @Darssal in #268

New Contributors

• @raashish1601 made their first contribution in #259
• @Thebearcoding made their first contribution in #257
• @Mrkinggggg made their first contribution in #256
• @Jah-yee made their first contribution in #255
• @kabishou11 made their first contribution in #254
• @jackiectl2 made their first contribution in #271
• @March-7 made their first contribution in #270
• @Darssal made their first contribution in #269
• @Ganesh1729-ui made their first contribution in #261

Full Changelog: v0.4.0...v0.4.1