Hugging Face's logo
Activity Feed

AI & ML interests

None defined yet.

Recent Activity
uyzhang  updated a dataset 3 days ago
Open-Bee/Bee-Training-Data-Stage1
uyzhang  updated a dataset 3 days ago
Open-Bee/Bee-Training-Data-Stage2
uyzhang  updated a dataset 3 days ago
Open-Bee/Honey-Data-1M
View all activity

Papers

Bee: A High-Quality Corpus and Full-Stack Suite to Unlock Advanced Fully Open MLLMs

View all Papers

Organization Card
Community About org cards

🐝 Open-Bee

Fully open data, curation pipeline and models for multimodal large language models.

🏠 Homepage · 📖 arXiv · 🤗 Collection · 💻 Code

🔥 Latest: Honey-Data-V2

Honey-Data-V2 is the corpus of the journal version of Bee, Bee: A High-Quality Corpus and Full-Stack Suite for End-to-End Training of Fully Open MLLMs, accepted to Computational Visual Media. It reapplies our HoneyPipe pipeline to Honey-Data-15M at a much larger scale: 44.3M samples comprising 100.3M QA pairs over 19.7M image groups, with long CoT in 17.9% of the samples.

  • Re-curated: the Honey-Data-15M pool is cleaned again under stricter structural rules, acting on community defect reports.
  • Multi-model rollout: two further independent rollouts by an upgraded model stack (Qwen3-VL-235B-A22B, Doubao-Thinking, Kimi-K2.5). Up to three verified answers are kept per instruction.
  • New sources: FineVision datasets absent from Honey-Data-15M, plus Molmo2-SynMultiImageQA, ChartVerse and Aria-UI, all passed through the same pipeline.

📦 Which dataset should I use?

dataset size what it is
Honey-Data-V2 🆕 44.3M samples Latest and largest SFT corpus (journal version). Not yet used to train the released Bee-8B.
Honey-Data-15M 15M samples SFT corpus of the conference version, with dual-level CoT. Bee-8B Stage 3.
Honey-Data-1M 1M samples Curated subset of Honey-Data-15M. Bee-8B Stage 4 (refinement SFT).
Bee-Training-Data-Stage1 – Bee-8B Stage 1 (MLP warm-up).
Bee-Training-Data-Stage2 – Bee-8B Stage 2 (vision-language alignment).

For the RL stage, merge ViRL39K and MMK12.

🤖 Models

model stage
Bee-8B-RL Final model, after RL
Bee-8B-SFT After supervised fine-tuning
Bee-8B-Stage1 · Stage2 · Stage3 Intermediate checkpoints

All released Bee-8B checkpoints were trained on Honey-Data-15M.

📰 News

  • [2026.09] 🍯 Honey-Data-V2 released. The journal version of Bee is accepted to Computational Visual Media.
  • [2026.03] 💻 Main code open-sourced.
  • [2026.01] 🎉 Bee accepted to ICLR 2026.
  • [2025.12] 🔥 All data and model weights across training stages released.
  • [2025.11] 📊 Honey-Data-15M and Honey-Data-1M released.
  • [2025.10] 🐝 Bee-8B released.

📝 Citation

@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},
}

@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},
}

Collections 1