MamMoth VL Instruct 12M
Mammoth-VL-Instruct-12M is a massive multimodal dataset that includes visual and textual instructions for mathematical and scientific problems, with detailed, step-by-step answers.
Approximately 37 million examples, 5 GB minimum, Parquet format
Apache 2.0
Description
The dataset Mammoth-VL-Instruct-12M offers a huge collection of sample visual instructions, with detailed answers, covering the fields of mathematics and science. Each example includes an image, an instruction, and a step-by-step solution, making it easy to train models with visual and logical reasoning.
What is this dataset for?
- Train multimodal models capable of understanding and solving complex visual problems
- Train AIs to provide detailed explanatory answers for scientific and mathematical questions
- Testing the visual and instructional reasoning ability of multimodal LLMs models
Can it be enriched or improved?
It is possible to enrich the dataset with additional annotations, to add more varied instructions or human corrections to improve the quality of the answers.
🔎 In summary
🧠 Recommended for
- Multimodal AI researchers
- Visual instruction model developers
- STEM AI projects
🔧 Compatible tools
- Hugging Face
- PyTorch
- TensorFlow
- Multimodal LLMs
💡 Tip
Use suitable batches to effectively manage this very large dataset.
Frequently Asked Questions
What are the disciplines covered by this dataset?
Mostly math and science with visual instructions and step-by-step solutions.
Can this dataset be used for commercial fine-tuning?
Yes, under the Apache 2.0 license, commercial use authorized with respect to the terms.
What size resources are required to work with this dataset?
It is recommended to have significant storage and calculation resources to fully exploit the 37 million examples.
