FT-LLM 2026 QA: Japanese Visual Question Answering Dataset for VLM Tuning

A Japanese visual-question-answering dataset used for Stage 1-2 visual instruction tuning of the COMPASS Vision-Language Model. Each sample contains a document or natural image together with one or more Japanese question–answer pairs. The dataset was created by Yana and last updated on 2026-04-16.

Use Cases

• Fine-tuning vision-language models for Japanese instruction-following based on the described question-answer pairs.
• Training models on Japanese visual question answering using the embedded document and natural images.
• Developing multimodal AI capabilities for Japanese language tasks based on the described visual instruction tuning purpose.

Strengths

• Images are embedded in the dataset, so no external downloads are required.
• Designed specifically for Stage 1-2 visual instruction tuning of a named VLM, indicating a clear purpose.

Limitations

• Row count, file formats, and column-level documentation are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Likely curated for training the COMPASS Vision-Language Model.

Freshness

Last updated 2026-04-16 13:33:29; freshness should be verified.