LLMsForHepth/llm_scores_hep_th

该数据集最初是LLMsForHepth/infer_hep_th的副本，随后新增了`score_Llama-3.1-8B`、`score_s1-L-3.1-8B-base`、`score_s2-L-3.1-8B-base`和`score_s3-L-3.1-8B-base_v3`列。这些新列中的值是真实摘要与不同模型生成的摘要之间的余弦相似度分数。具体来说，`score_Llama-3.1-8B`中的相似度分数是通过使用`meta-llama/Llama-3.1-8B`基础模型对`abstract`和`comp_Llama-3.1-8B`列中的序列进行嵌入计算得出的，其余分数则是使用微调模型进行嵌入计算的。嵌入计算采用均值池化方法。

This dataset originates as a copy of LLMsForHepth/infer_hep_th, with additional columns added to store cosine similarity scores between the ground truth abstracts and the abstracts generated by different models. These scores are calculated by embedding the ground truth abstracts and the generated abstracts and then computing the similarity.