4M Img Caps: 4 Million Image-Caption Pairs

4,000,000 image-caption pairs stored in PyArrow IPC format for high-performance multimodal training. The dataset utilizes memory-mapped files to enable low-latency data access during large-scale model optimization.

Use Cases

• Train multimodal encoders using the image and caption features
• Benchmark I/O performance using the pa.memory_map and pa.ipc.open_file methods
• Fine-tune generative models on 4 million text-image pairs

Strengths

• 4,000,000 image-caption pairs
• Distributed as PyArrow IPC files for memory-mapped reading
• Includes a PyTorch Dataset class template for immediate integration