Add dataset card and metadata

README.md

@@ -0,0 +1,23 @@
+---
+license: mit
+task_categories:
+  - image-text-to-text
+tags:
+  - visual-reasoning
+  - data-efficient
+  - mcts
+---
+
+# ThinkLite-VL Datasets
+
+This repository contains two datasets used in the paper "SoTA with Less: MCTS-Guided Sample Selection for Data-Efficient Visual Reasoning Self-Improvement":
+
+* **ThinkLite-VL-70k:** A large-scale dataset.  [Hugging Face Dataset](https://huggingface.co/datasets/russwang/ThinkLite-VL-70k)
+* **ThinkLite-VL-hard-11k:** A smaller, more challenging subset of the data. [Hugging Face Dataset](https://huggingface.co/datasets/russwang/ThinkLite-VL-hard-11k)
+
+
+These datasets are designed for data-efficient visual reasoning and are used to evaluate the performance of models trained with MCTS-guided sample selection.
+
+**Paper:** [SoTA with Less: MCTS-Guided Sample Selection for Data-Efficient Visual Reasoning Self-Improvement](https://arxiv.org/abs/2504.07934)
+
+**Associated Model:** [ThinkLite-VL-7B](https://huggingface.co/russwang/ThinkLite-VL-7B)