yosubshin/walton-hard-mm-mathinstruct-mmr1

walton-hard-mm-mathinstruct-mmr1数据集是一个用于视觉语言模型训练的数据集，它结合了 Walton 和 MMR1 数据集。该数据集经过特定的预处理配置，包括使用Qwen/Qwen2.5-7B-Instruct作为基础模型和分词器，序列长度为16384，处理类型为视觉语言。数据集特征包括输入序列、注意力掩码、语言模型标签、PIL图像对象、原始对话消息和处理元数据。该数据集共有10000个原始样本和经过处理的样本，处理成功率为100%。

The walton-hard-mm-mathinstruct-mmr1 dataset is a vision-language model training dataset that combines the Walton and MMR1 datasets. This dataset has undergone specific preprocessing configurations, including using Qwen/Qwen2.5-7B-Instruct as the base model and tokenizer, with a sequence length of 16384 and a processing type of vision language. The dataset features include input sequences, attention masks, language model labels, PIL Image objects, original conversation messages, and processing metadata. The dataset consists of 10000 original and processed samples with a success rate of 100%.