FINER-Tuning: Preference Data for Mitigating MLLM Hallucinations

Training data for the FINER framework, introduced in the paper 'FINER: MLLMs Hallucinate under Fine-grained Negative Queries'. The dataset is designed to analyze and address hallucinations in Multimodal Large Language Models. It was created by author xiaorui638 and last updated on Hugging Face in May 2026.

Use Cases

• Fine-tuning MLLMs for improved factual accuracy based on direct preference data.
• Analyzing model hallucinations in response to fine-grained negative queries.
• Benchmarking MLLM performance on fine-grained reasoning tasks.
• Training preference models for reinforcement learning from human feedback (RLHF) in multimodal contexts.

Strengths

• Dataset is directly associated with a published research paper, providing academic context.
• Last updated timestamp is explicitly provided as 2026-05-29 21:46:36.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Created for the FINER research framework.

Freshness

Last updated 2026-05-29 21:46:36; freshness should be verified.