知识卡片

普通读书笔记卡 · 8700.f · 生成模型:GPT-6

缓存对比损失换取有效大批次

理解等级:高级

缓存式训练可降低一次性保留全部计算图的内存压力,让更多样本共同参与对比。它改变训练执行方式,不自动修复错标签;是否更快和更准确仍取决于批内负例质量及实现开销。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Use CachedMultipleNegativesRankingLoss, available in Sentence Transformers, which allows you to use larger batch sizes, leading to better performance. Now that we have discussed all the ingredients needed for fine-tuning, let’s put it all together with the SentenceTransformerTrainer class: from datasets import load_dataset from sentence_transformers import SentenceTransformer, SentenceTransformerTrainer from sentence_transformers.losses import TripletLoss model = SentenceTransformer( "'all-mpnet-base-v2'") dataset = load_dataset("csv", data_files="negatives_dataset.csv") loss = TripletLoss(model) trainer = SentenceTransformerTrainer(

    参考来源

    - 位置:《Designing Large Language Model Applications》第11章《表示学习与嵌入》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch11.html) - 结论依据:摘录中关于缓存对比损失换取有效大批次的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Use CachedMultipleNegativesRankingLoss, available in Sentence Transformers, which allows you to use larger batch sizes, leading to better performance. Now that we have discussed all the ingredients needed for fine-tuning, let’s put it all together with the SentenceTransformerTrainer class: from datasets import load_dataset from sentence_transformers import SentenceTransformer, SentenceTransformerTrainer from sentence_transformers.losses import TripletLoss model = SentenceTransformer( "'all-mpnet-base-v2'") dataset = load_dataset("csv", data_files="negatives_dataset.csv") loss = TripletLoss(model) trainer = SentenceTransformerTrainer(