Contamination Map Methodology

Text / LLM / Model

LLaVA-OneVision

Organisation
LLaVA-NeXT team (LMMs-Lab), with ByteDance
Family
LLaVA
Released
2024-08-06 (arXiv v1 submission date (2408.03326))
Checkpoints
lmms-lab/llava-onevision-qwen2-0.5b-si, lmms-lab/llava-onevision-qwen2-0.5b-ov, lmms-lab/llava-onevision-qwen2-7b-si, lmms-lab/llava-onevision-qwen2-7b-ov, lmms-lab/llava-onevision-qwen2-72b-si, lmms-lab/llava-onevision-qwen2-72b-ov-sft, lmms-lab/llava-onevision-qwen2-7b-ov-chat, lmms-lab/llava-onevision-qwen2-72b-ov-chat
Weights
Open · Apache 2.0
Links
PaperModel cardCode

Training data disclosure: Partial list

Level determined for 18% of benchmark cells.

“We list the full details about each dataset in our collection, and their categorization and formatting prompt in Appendix E.3.”
arxiv.org · paper · Section 4.2 · retrieved 2026-10-15
“The dataset statistics are summarized in Table 16.”
arxiv.org · paper · Appendix E.3 · retrieved 2026-10-15

Built on

Everything these base models may have seen is inherited by LLaVA-OneVision.

Other declared training data

Benchmark cells

Classic LLM benchmarks

Unknown MMLU (Measuring Massive Multitask Language Understanding) · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown GSM8K (Grade School Math 8K) · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown MATH (Mathematics Aptitude Test of Heuristics) · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown HumanEval · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown ARC-Challenge (AI2 Reasoning Challenge, Challenge Set) · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown HellaSwag · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown TruthfulQA · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown WinoGrande · unknown for its base model SigLIP: training data not disclosed in enough detail

Multimodal LLM benchmarks

Unknown MMMU (Massive Multi-discipline Multimodal Understanding and Reasoning) · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown MMMU-Pro · unknown for its base model SigLIP: training data not disclosed in enough detail

Related data seen MathVista · in fine-tuning data: related dataset ChartQA was used

Evidence (6)
“† indicates that the training set has been observed in our data mixture.”
arxiv.org · paper · Table 2 caption · retrieved 2026-10-15
“Single-Image †ChartQA [101]”
arxiv.org · paper · Table 2 · retrieved 2026-10-15
“ChartQA (18.3 K)”
arxiv.org · paper · Figure 4 (Single-Image 3.2M) · retrieved 2026-10-15
“It consists of 6,141 examples, derived from 28 existing multimodal datasets involving mathematics and 3 newly created datasets”
arxiv.org · paper · Abstract · retrieved 2026-10-15
“ChartQA (2022)”
arxiv.org · paper · Table 5 (Appendix C) · retrieved 2026-10-15
“"dataset": "ChartQA",”
huggingface.co · dataset card · source.json · retrieved 2026-10-15

Training data seen ChartQA · in fine-tuning data: only the training split was used, not the benchmark's test data

Evidence (3)
“† indicates that the training set has been observed in our data mixture.”
arxiv.org · paper · Table 2 caption · retrieved 2026-10-15
“Single-Image †ChartQA [101]”
arxiv.org · paper · Table 2 · retrieved 2026-10-15
“ChartQA (18.3 K)”
arxiv.org · paper · Figure 4 (Single-Image 3.2M) · retrieved 2026-10-15

Training data seen DocVQA · in fine-tuning data: only the training split was used, not the benchmark's test data

Evidence (3)
“† indicates that the training set has been observed in our data mixture.”
arxiv.org · paper · Table 2 caption · retrieved 2026-10-15
“†DocVQA [103] (test)”
arxiv.org · paper · Table 2 · retrieved 2026-10-15
“Doc-VQA (10.2 K)”
arxiv.org · paper · Figure 4 (Single-Image 3.2M) · retrieved 2026-10-15

Training data seen AI2D (AI2 Diagrams) · in fine-tuning data: only the training split was used, not the benchmark's test data

Evidence (4)
“† indicates that the training set has been observed in our data mixture.”
arxiv.org · paper · Table 2 caption · retrieved 2026-10-15
“†AI2D [53]”
arxiv.org · paper · Table 2 · retrieved 2026-10-15
“AI2D (InternVL) (12.4 K)”
arxiv.org · paper · Figure 4 (Single-Image 3.2M) · retrieved 2026-10-15
“we used images from the AI2D and InfoVQA training sets and employed GPT-4V to provide detailed descriptions of the images”
arxiv.org · paper · Appendix E.1 · retrieved 2026-10-15

Unknown CharXiv · training data not disclosed in enough detail

Recent LLM benchmarks

Unknown GPQA Diamond · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown MMLU-Pro · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown LiveCodeBench (code generation, release_v6) · training data not disclosed in enough detail

Unknown SWE-bench Verified · unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown AIME 2024 · training data not disclosed in enough detail

Unknown AIME 2025 · training data not disclosed in enough detail

Unknown MATH-500 · unknown for its base model SigLIP: training data not disclosed in enough detail

Notes

The authors mark AI2D, ChartQA, DocVQA (and InfoVQA) as in-domain because their training sets are in the data mixture. The data mix was tuned while monitoring academic benchmarks incl. AI2D, MMMU and MathVista (Appendix E.1). Third-party: the FineVision study (Hugging Face) found 2.15% of the LLaVA-OneVision instruction-data images to be near-duplicates (SSCD cosine >= 0.95) of images from 66 lmms-eval test sets, with a 2.7 pp average score drop after removing them (FineVision Table 4); per-benchmark numbers are only in a figure, so this is not recorded as a per-dataset finding. ov-chat checkpoints have additional preference training. Accepts image input: “Three vision scenarios are considered, depding which visual input is considered in the multimodal sequence, including single-image, multi-image, video.”

Sources