KomeijiForce/Cuckoo_C4_1M

Cuckoo是一个小型信息提取模型，通过标记给定输入上下文中的下一个标记来预测下一个标记。它可以利用任何文本资源进行增强，特别是利用为大规模语言模型定制的数据。README中提供了Cuckoo在CoNLL2003、BioNLP2004、MIT-Restaurant、MIT-Movie等任务上的性能表现，以及如何使用Cuckoo进行零样本提取、少样本适应和自定义预训练的示例代码和脚本。Cuckoo is a small information extraction model that predicts the next tokens by tagging them in the given input context. It can enhance itself with any text resource, especially leveraging data curated for large language models. The README provides performance metrics of Cuckoo on tasks like CoNLL2003, BioNLP2004, MIT-Restaurant, and MIT-Movie, along with example code and scripts for zero-shot extraction, few-shot adaptation, and custom pre-training.