--- license: other license_name: mixed-see-licensing-information task_categories: - visual-question-answering - image-text-to-text language: - en - zh tags: - multimodal - vision-language - instruction-tuning - chain-of-thought - sft - ocr - chart - document - grounding - gui size_categories: - 10M The released Bee-8B checkpoints, and all results reported in the paper, were trained on Honey-Data-15M. > Evaluating models trained on Honey-Data-V2 is left for future work. ## What is new compared with Honey-Data-15M Honey-Data-V2 is not built from scratch. It extends Honey-Data-15M in three ways, all executed with the existing operators of DataStudio: - **Re-curation of the existing pool.** Acting on community defect reports, samples whose final turn carries no answer were removed, and 492.6K grounding instructions were rewritten to state their coordinate convention up front. 1.8M machine-recaptioning samples were dropped, because they dominated by volume while adding little instruction-following signal. Image placeholders were canonicalized, and degenerate generations (unpaired reasoning tags, repetition collapse, placeholder leakage) were rejected. - **Multi-model rollout.** On top of the original annotation pass, two further independent rollouts answered the same instructions with an upgraded model stack. Every verified answer is kept, so an instruction carries one, two or three independently produced answers. Over the overlapping subset, two answers to the same instruction agree verbatim in only 0.00%–0.89% of cases. - **New source incorporation.** The [FineVision](https://huggingface.co/datasets/HuggingFaceM4/FineVision) datasets absent from Honey-Data-15M, together with Molmo2-SynMultiImageQA, ChartVerse and Aria-UI, were passed through the same pipeline rather than concatenated as-is. Sources whose answers are already determinate, such as OCR transcription and box-level grounding, were filtered but not rewritten. | lineage | samples | long CoT | |---|---:|---:| | re-curated Honey-Data-15M pool (rollout passes 1 / 2 / 3: 9.6M / 12.8M / 7.0M) | 29.4M | 15.5% | | newly incorporated sources | 14.9M | 22.6% | | **Honey-Data-V2 total** | **44.3M** | **17.9%** | | role | Honey-Data-15M | Honey-Data-V2 | |---|---|---| | short CoT | Qwen2.5-VL-72B/32B | Qwen3-VL-235B-A22B-Instruct, Kimi-K2.5 | | long CoT | proprietary MLLMs | Qwen3-VL-235B-A22B-Thinking, Doubao-Thinking, Kimi-K2.5 | | verification | Qwen2.5-VL-72B | Qwen3-235B-A22B-Instruct | | GUI grounding | – | MAI-UI-8B | ## One record is one image group Unlike the usual one-row-per-conversation layout, a record here collects **every conversation that refers to the same image**. The image bytes are therefore stored once, no matter how many dialogues discuss it. Across the corpus this is the difference between 19,708,047 stored images and the 44,295,078 conversations that use them. ```python from datasets import load_dataset, get_dataset_config_names subsets = get_dataset_config_names("HoneyDataV2/Honey-Data-V2") ds = load_dataset("HoneyDataV2/Honey-Data-V2", subsets[0], split="train") rec = ds[0] rec["images"][0] # a PIL image len(rec["data"]) # how many dialogues share it rec["data"][0]["messages"] # [{'from': 'human', 'value': ...}, {'from': 'gpt', 'value': ...}] rec["data"][0]["provenance"] # which model wrote and which verified this answer ``` ## Image encoding Images are stored as **lossless WebP**. Where the source was already JPEG, or where WebP came out larger than the original, the original bytes are kept untouched. WebP holds the same pixels as the PNGs it replaces in roughly 60% of the bytes. Decoding is transparent: `datasets` returns PIL images either way. For RGBA images the alpha channel is preserved exactly and RGB is preserved wherever alpha > 0. PNG stores arbitrary colour under fully transparent pixels and WebP does not carry it over; no reader can observe those values. ### Record fields | column | type | meaning | |---|---|---| | `id` | string | identifier of the image group | | `images` | list of image | the images, decoded to PIL by `datasets` | | `n_images` | int32 | how many images the group holds | | `image_sizes` | list of [width, height] | pixel dimensions, one pair per image | | `image_phashes` | list of string | perceptual hash, one per image | | `category` | string | task category; matches the directory | | `source_subset` | string | subset; matches the directory | | `categories` | list of string | every category label the upstream corpora gave this group | | `sources` | list of string | every upstream corpus that contributed a dialogue | | `source_subsets` | list of string | every upstream subset name | | `data` | list of struct | the dialogues, below | Each element of `data`: | field | meaning | |---|---| | `id`, `ori_id` | identifiers of this dialogue and of the record it came from | | `messages` | the dialogue as `from` / `value` pairs, `human` and `gpt` | | `n_turns` | number of turns | | `cot_level` | `short` or `long` chain-of-thought | | `original_answers` | the answers before rewriting | | `provenance` | which model performed each step | | `source`, `subset`, `source_subset`, `category` | where this dialogue came from | `provenance` names the model used at each stage, so a subset of the corpus can be selected by annotator: `question_image_consist` (the image-question consistency check), `answer_rewrite`, `reasoning_rewrite`, and `answer_consist` (verification of the rewritten answer). A value of `null` means that stage did not apply — determinate tasks such as OCR transcription and box-level grounding are filtered but never rewritten. ## Composition | category | subsets | image groups | conversations | size | |---|---:|---:|---:|---:| | Caption | 13 | 1,413,122 | 2,011,253 | 1.21 TB | | Chart | 77 | 4,665,363 | 11,036,064 | 0.87 TB | | Document | 49 | 2,271,245 | 4,766,060 | 0.67 TB | | GUI-Grounding | 4 | 500,916 | 2,803,848 | 0.12 TB | | General | 122 | 4,298,102 | 12,952,088 | 1.75 TB | | Grounding | 27 | 1,979,196 | 3,435,355 | 0.75 TB | | OCR | 45 | 2,564,793 | 3,625,483 | 0.37 TB | | STEM | 83 | 2,015,115 | 3,664,927 | 0.07 TB | | **total** | **420** | **19,707,852** | **44,295,078** | **5.81 TB** | A subset name that appears in more than one category is qualified with its category, for example `Chart-CoSyn` and `Document-CoSyn`. ## Provenance of the answers | stage | models | |---|---| | answer rewriting | Qwen3-VL-235B-A22B-Instruct, Kimi-K2.5, MAI-UI-8B | | reasoning rewriting | Qwen3-VL-235B-A22B-Thinking, Kimi-K2.5, Doubao-Thinking | | image-question consistency | Qwen3-VL-30B-A3B-Instruct | | answer verification | Qwen3-235B-A22B-Instruct | `sources` records which annotation pass a dialogue belongs to: `Honey-Data-15M-Rollout1` is the re-curated original pool, `Rollout2` and `Rollout3` are the two further independent passes over the same instructions. Dialogues taken from newly incorporated community corpora carry that corpus's name instead. ## Licensing information Honey-Data-V2 aggregates many publicly available datasets, each governed by its own licence. Anyone using this corpus must observe the licence attached to each constituent dataset. To the extent we hold any rights in the curation and the generated answers, those are released under CC-BY-4.0. ## Citation If you use Honey-Data-V2, please cite the journal version: ```bibtex @article{zhang2026beecvm, title = {Bee: A High-Quality Corpus and Full-Stack Suite for End-to-End Training of Fully Open MLLMs}, author = {Yi Zhang and Bolin Ni and Xin-Sheng Chen and Heng-Rui Zhang and Yongming Rao and Houwen Peng and Qinglin Lu and Han Hu and Meng-Hao Guo and Shi-Min Hu}, journal = {Computational Visual Media}, year = {2026}, note = {to appear}, } ``` The conference version, which introduced Honey-Data-15M and Bee-8B: ```bibtex @inproceedings{zhang2026bee, title = {Bee: A High-Quality Corpus and Full-Stack Suite to Unlock Advanced Fully Open MLLMs}, author = {Yi Zhang and Bolin Ni and Xin-Sheng Chen and Heng-Rui Zhang and Yongming Rao and Houwen Peng and Qinglin Lu and Han Hu and Meng-Hao Guo and Shi-Min Hu}, booktitle = {The Fourteenth International Conference on Learning Representations}, year = {2026}, } ```