freedomintelligence / dptdr Goto Github PK

How to pre-train some parameters of deep prompt？
Pre-train deep prompts One is to pre-train deep prompts with a vanilla PLM. Later we initialize a DPT-based retriever using the pre-trained deep prompts and the vanilla PLM. However, experiments in Sec. 4.4 show that it suffers from catastrophic forgetting and exhibits no superior performance to randomly initialized prompts.
how to get the "pre-trained deep prompts"?

Hello, I plan to run your code, but I found that the versions of the packages you provided do not match; for example, transformers=4.9.2, datasets=1.17.0. These two versions conflict. Can you provide the complete environment for all the code, such as the PyTorch version, CUDA version, etc.?
Thanks in advance!
Yy

看到DPTDR在百度段落检索上拿到了top1，请问你们有在百度段落检索的数据集上跟其他算法做对比吗。