ETCHR SFT-400K: 400,000 Samples for Visual Reasoning Assistant Training

400,000 samples across five tasks were used to transfer a passive image editor into an autonomous, question-conditioned visual reasoning assistant. The dataset was created by BeichenZhang and last updated on 2026-05-25. It includes tasks such as Fine-grained Perception, Chart Understanding, Maze Solving, and Jigsaw Puzzle.

Use Cases

• Fine-tuning vision-language models for fine-grained perception tasks based on the described task categories.
• Training autonomous visual reasoning assistants based on the question-conditioned samples.
• Benchmarking model performance on chart understanding and data interpretation tasks.
• Developing agents capable of solving maze and puzzle navigation problems based on visual input.

Strengths

• Contains 400,000 training samples, providing a substantial base for model fine-tuning.
• Designed for a specific, complex task: transferring a passive image editor into an autonomous visual reasoning assistant.
• Covers five distinct task types, including Fine-grained Perception and Chart Understanding.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-05-25 11:45:27; freshness should be verified.

Provenance

Source

BeichenZhang on Hugging Face

Collection Method

Created as SFT (Supervised Fine-Tuning) training data for the ETCHR project.

Freshness

2026-05-25