InternVideo3: Long-Video QA Annotations for Supervised Fine-Tuning

Yanziang's InternVideo3 dataset provides 380,000 samples for supervised fine-tuning of models on long-video understanding. Each sample contains a YouTube video ID paired with question-answer annotations for detailed description and reasoning tasks. The dataset was uploaded to Hugging Face and last updated on June 11, 2026.

Use Cases

• Fine-tuning video-language models for detailed video description based on the provided QA annotations.
• Training models for complex reasoning about long-video content based on the described annotations.
• Benchmarking agent capabilities in multimodal contextual reasoning tasks using the structured QA pairs.

Strengths

• Contains 380,000 samples for model training and evaluation.
• Annotations are structured for detailed description and reasoning tasks, as mentioned in the description.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face dataset uploaded by author yanziang.

Collection Method

Likely contains YouTube video IDs paired with manually or synthetically generated QA annotations.

Freshness

Last updated 2026-06-11 02:45:59; freshness should be verified.