ytu-ce-cosmos/Turkish-LLaVA-Pretrain

TurkishLLaVA预训练数据集是一个土耳其语翻译版本的数据集，原始数据集是之前研究中使用的英文数据集。该数据集使用DeepL进行翻译，以保持原始数据的细微差别和上下文。在预训练阶段，模型只学习解释图像编码器的输出，专注于如何将视觉信息与语言模型对齐。因此，模型在此阶段还不能进行对话或处理特定任务的查询。

The TurkishLLaVA Pretrain Dataset is a Turkish translation of the English dataset used in previous studies. The translation was carefully executed using DeepL to preserve the nuances and context of the original data. In this pretraining phase, the model only learns to interpret the output of the image encoder, focusing on how to align visual information with the language model. As a result, the model is not yet capable of engaging in conversations or handling task-specific queries.