Object-hallucination probe trained on a LLaVA-1.5-13B COCO corpus, with three independent training seeds (0, 42, 1337). It is not interchangeable with the 7B release: features, geometry and thresholds are model-specific. 10,000 COCO train2014 images split 80/20 by image: 8,000 train / 54,253 object spans and 2,000 holdout / 13,789 object spans. Captions are generated by 13B itself, so this corpus is not the 7B corpus. Thresholds are selected by best F1 on the COCO validation split and then frozen. Nothing is tuned on the test holdout, AMBER, or POPE. COCO test holdout — mention view, n=13,787, 16.19% positive AMBER, strict zero-shot — n=2,367 rows / 977 images, 5.37% positive, COCO…
Independent publisher
Đinh Quốc Bảo
ToiTenBao
Models in Library7
Datasets in Library1
Models on Hugging Face13
Followers—
Compare the published evidence
This profile connects the 8 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Image and text to text.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
Models
Object-hallucination probe trained on a LLaVA-1.5-13B COCO corpus, with three independent training seeds (0, 42, 1337). It is not interchangeable with the 7B release: features, geometry and thresholds are model-specific. 10,000 COCO train2014 images split 80/20 by image: 8,000 train / 54,253 object spans and 2,000 holdout / 13,789 object spans. Captions are generated by 13B itself, so this corpus is not the 7B corpus. Thresholds are selected by best F1 on the COCO validation split and then frozen. Nothing is tuned on the test holdout, AMBER, or POPE. COCO test holdout — mention view, n=13,787, 16.19% positive AMBER, strict zero-shot — n=2,367 rows / 977 images, 5.37% positive, COCO…
Object-hallucination probe trained on a LLaVA-1.5-13B COCO corpus, with three independent training seeds (0, 42, 1337). It is not interchangeable with the 7B release: features, geometry and thresholds are model-specific. 10,000 COCO train2014 images split 80/20 by image: 8,000 train / 54,253 object spans and 2,000 holdout / 13,789 object spans. Captions are generated by 13B itself, so this corpus is not the 7B corpus. Thresholds are selected by best F1 on the COCO validation split and then frozen. Nothing is tuned on the test holdout, AMBER, or POPE. COCO test holdout — mention view, n=13,787, 16.19% positive AMBER, strict zero-shot — n=2,367 rows / 977 images, 5.37% positive, COCO…
Object-hallucination probe trained on a LLaVA-1.5-13B COCO corpus, with three independent training seeds (0, 42, 1337). It is not interchangeable with the 7B release: features, geometry and thresholds are model-specific. 10,000 COCO train2014 images split 80/20 by image: 8,000 train / 54,253 object spans and 2,000 holdout / 13,789 object spans. Captions are generated by 13B itself, so this corpus is not the 7B corpus. Thresholds are selected by best F1 on the COCO validation split and then frozen. Nothing is tuned on the test holdout, AMBER, or POPE. COCO test holdout — mention view, n=13,787, 16.19% positive AMBER, strict zero-shot — n=2,367 rows / 977 images, 5.37% positive, COCO…
Object-hallucination probe trained on a LLaVA-1.5-13B COCO corpus, with three independent training seeds (0, 42, 1337). It is not interchangeable with the 7B release: features, geometry and thresholds are model-specific. 10,000 COCO train2014 images split 80/20 by image: 8,000 train / 54,253 object spans and 2,000 holdout / 13,789 object spans. Captions are generated by 13B itself, so this corpus is not the 7B corpus. Thresholds are selected by best F1 on the COCO validation split and then frozen. Nothing is tuned on the test holdout, AMBER, or POPE. COCO test holdout — mention view, n=13,787, 16.19% positive AMBER, strict zero-shot — n=2,367 rows / 977 images, 5.37% positive, COCO…
Datasets
Canonical COCO and AMBER detection packages are under detection/ /models/ /probes/ /. See detection/index.json for active and pending packages. Each published package has a manifest with exact checkpoint, cache and test-result paths. Seeds are 0, 42 and 1337. TruthPrInt retains its native validation TPR-at-1%-FPR checkpoint; other detection probes use best pooled validation AUROC. Detection thresholds are validation-best F1. COCO and AMBER are independent training/evaluation datasets. COCO 13B Ours/VIB checkpoints, caches and results are copied here; their frozen source records are not yet published, so these two packages remain pending in the index. ablation/ contains separately indexed…