Add task categories, link to paper, and Github repository

This PR ensures the dataset is linked to (and can be found at) https://huggingface.co/papers/2410.13924.
It also adds the `image-segmentation` task category and links to the Github repository.

README.md

@@ -1,21 +1,17 @@
 ---
-viewer: false
-license: bsd
+base_model:
+- meta-llama/Meta-Llama-3-8B
+datasets:
+- QizhiPei/MathFusionQA
 language:
 - en
-tags:
-- 3D semantic segmentation
-- indoor 3D scene dataset
-pretty_name: arkit_labelmaker
-size_categories:
-- 1K<n<10K
+pipeline_tag: question-answering
+library_name: transformers
+license: mit
 ---
 
-# ARKit Labelmaker: A New Scale for Indoor 3D Scene Understanding
+Paper: [MathFusion: Enhancing Mathematic Problem-solving of LLM through Instruction Fusion](https://arxiv.org/abs/2503.16212)
 
-[[arxiv]](https://arxiv.org/abs/2410.13924)  [[website]](https://labelmaker.org/) [[checkpoints]](https://huggingface.co/labelmaker/PTv3-ARKit-LabelMaker)
+MathFusion enhances mathematical reasoning through cross-problem instruction synthesis, implementing sequential, parallel, and conditional fusion strategies.
 
-We complement ARKitScenes dataset with dense semantic annotations that are automatically generated at scale. This produces the first large-scale, real-world 3D dataset with dense semantic annotations.
-Training on this auto-generated data, we push forward the state-of-the-art performance on ScanNet and ScanNet200 with prevalent 3D semantic segmentation models.
-
-              
+Code: https://github.com/QizhiPei/MathFusion