penfever/vlaa-thinking-sft-to-r1-format

vlaa-thinking-sft-to-r1-format

vlaa-thinking-sft dataset transformed to match multimodal-open-r1-8k-verified format with filtering

Dataset Description

This dataset was processed using the data-preproc package for vision-language model training.

Processing Configuration

• —Base Model: allenai/Molmo-7B-O-0924
• —Tokenizer: allenai/Molmo-7B-O-0924
• —Sequence Length: 8192
• —Processing Type: Vision Language (VL)

Dataset Features

• —input_ids: Tokenized input sequences
• —attention_mask: Attention masks for the sequences
• —labels: Labels for language modeling
• —images: PIL Image objects
• —messages: Original conversation messages
• —metadata: Processing metadata

Processing Statistics

• —Original Samples: 12483
• —Processed Samples: 12483
• —Success Rate: 100.0%
• —Average Token Length: N/A
• —Max Token Length: N/A
• —Truncation Rate: N/A

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("your-org/your-dataset-name")

# Access samples
sample = dataset["train"][0]
print(f"Input tokens: {len(sample['input_ids'])}")
print(f"Images: {len(sample['images'])}")
print(f"Messages: {sample['messages']}")

Citation

If you use this dataset in your research, please cite:

@misc{your-dataset-name,
  title={Your Dataset Name},
  author={Your Name},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/your-org/your-dataset-name}
}

License

This dataset is released under the specified license. Please check the license field for details.