llava-15-7b-captioning-restricting-lora-adapter-attention-heads-picked-hallucination CLM-0076In LLaVA-1.5-7B captioning, restricting a LoRA adapter to attention heads picked by a hallucination diagnosis (attention-to-image drop around hallucinated object words, then ablation screening) reduces object hallucination, while an identically trained layer-matched random-head adapter does not — so the selection, not the added adapter capacity, carries the effect.
Capability: Grounding answers in the image
Observed on
One model (LLaVA-1.5-7B), one captioning prompt, COCO val2014 images, CHAIR scoring; 32 selected heads across 19 layers; control split of 200 images. Gains come with shorter captio.
Sources
- Measured on 400 held-out COCO images (CHAIRs 0.370 to 0.230, CHAIRi 0.156 to 0.096, paired sign-flip p<0.001) plus a 200-image control split where random-head LoRA gave CHAIRs 0.400 vs 0.100 for selected heads. Fixed-budget runs (64/80/128 tokens) rule out pure truncation, but no fully length-matched or all-head-LoRA baseline, and recall falls, so precision-coverage tradeoff is not isolated.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- In LLaVA-1.5-7B captioning, restricting a LoRA adapter to attention heads picked by a hallucination diagnosis (attention-to-image drop around hallucinated object words, then ablation screening) reduces object hallucination, while an identically trained layer-matched random-head adapter does not — so the selection, not the added adapter capacity, carries the effect.Grounding answers in the image · reviewed by AI
- In 7B-class vision-language models (LLaVA-1.5, MiniGPT-4, mPLUG-Owl2), object hallucination is carried by an anisotropic set of directions in deep-layer hidden states, so attenuating each eigenmode in proportion to its hallucination-to-truthful variance ratio — estimated once offline from paired truthful/hallucinated captions and folded into the FFN output projection — lowers CHAIR rates, while a mean-offset subtraction or isotropic shrinkage of the same representations does not.Grounding answers in the image · reviewed by AI
- In 7B-class LVLMs (LLaVA-1.5, MiniGPT-4, mPLUG-Owl2), hallucination directions in activation space are not a single mode: clustering hallucinated-minus-faithful hidden-state differences into several low-rank subspaces and weighting them per test image lowers object-hallucination rates slightly below a single global null-space projection, with the margin small (CHAIR_I 5.30 to 4.92 on LLaVA-1.5) and inconsistent on yes/no existence probing.Grounding answers in the image · reviewed by AI
- In open-source vision-language models (LLaVA-1.5, Qwen-VL-Chat, Qwen2.5-VL, LLaVA-NeXT), a training-free decoding intervention that detects large layer-to-layer and step-to-step shifts in hidden states and nudges the diverging states back toward their prior value lowers object-hallucination rates, but the lower rate comes with fewer objects mentioned and shorter outputs.Grounding answers in the image · reviewed by AI
- In 4-7B open vision-language models (LLaVA-1.5-7B, Qwen2.5-VL-7B, InternVL-2.5-4B), how the cross-modal attention distribution over visual tokens shifts between adjacent decoder layers separates hallucinated from grounded object predictions better than the model's sensitivity to masking the attended region, and better than attention drift measured against a fixed first or last layer anchor.Grounding answers in the image · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A random-head or all-head LoRA trained with the same DPO pairs and matched length reaching the same CHAIR reduction, or the selected heads failing to transfer to other VLMs/prompts. Proposed technique, not catalogued: Diagnosis-selected head-restricted adapter plus grounding-gated decoding.