ByteDance-Seed / ByteDance-Seed/Bagel
Pretraining VLM datasets and Training parameters
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 6.2k
- Forks
- 545
- PR merge metrics
- No merged PRs in 30d
Description
Hi, thank you for open-sourced code!
I am planning to pretrain the Qwen2.5 0.5B/3B with NaViT over the general knowledge datasets (Actually, I found that the VLM part of Bagel cannot load the pre-trained Qwen2.5-VL because of the dimension mismatch, it seems that the VLM part of the Bagel is manually trained, am I right?).
I am wondering what kind of and what size of the datasets should I used for training (is BLIP558K used by LLaVA 1.5 sufficient?), and can you suggest the training parameters as well as estimated training time on 4 H100 gpus?
Many thx!
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
The issue names Qwen2.5, NaViT, BLIP558K, and training on four H100 GPUs, but it identifies no repository files, tests, or entry points. Locate the existing pretraining documentation and configuration entry points, then clarify which dataset guidance, parameters, and timing information should be documented.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- data, machine-learning
- Issue type
- Documentation
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100