Therascribe Gold 1M: Research-Backed Medical Vision-Language Dataset for Fine-Tuning

753,715 medical image-text pairs totaling 49.37 GB, designed for fine-tuning models like LLaVA-Med++. The dataset, created by Kafoo and last updated in November 2025, is stored in JSONL format alongside its images. Its captions are notably concise, averaging 1.0 words in length.

Use Cases

• Fine-tuning vision-language models like LLaVA-Med++ based on the described research-backed medical image-text pairs.
• Training medical image captioning systems based on the dataset's large collection of images with associated text.
• Benchmarking multimodal AI performance in the medical domain based on the dataset's scale and structure.
• Developing medical visual question answering systems based on the vision-language format implied by the description.

Strengths

• Contains 753,715 samples, indicating a substantial scale for training.
• Total size of 49.37 GB suggests high-resolution or numerous image files.
• Explicitly designed for research-backed medical vision-language tasks, providing a clear application focus.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The average caption length of 1.0 words suggests textual annotations may be very brief or limited.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Kafoo (author on Hugging Face).

Collection Method

Likely aggregated or created from research sources for model fine-tuning.

Time Range

null

Freshness

Last updated 2025-11-30 21:00:00; freshness should be verified.

Geography

null