-
Add self-contained TOSC evaluation pipeline
julyanghar +1: Claude Opus 4.8 - eval/: scoring engine (eval_tosc.py), synonym table, build_index.py
(image folders -> TOSC_dataset.jsonl + insertions.jsonl), run_eval.sh
- llava/: vendored inference package (model_vqa + LLaVA-1.5 loader + Qwen
helpers) so evaluation runs with no external dependency
- benchmark/: generated index files (5969 triplets)
- requirements.txt + READMEs (install & run guide)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>