Runtime error when loading a pretrained model, ProQA (20 prompts+paq)

[starsuzi]

Hello,

Thanks for sharing such an exciting work!

However, I am facing an error when loading the pretrained model, ProQA (20 prompts+paq).
Specifically, I met a Runtime error when I ran a zero-shot experiment, as follows:
RuntimeError: Error(s) in loading state_dict for T5ForConditionalGeneration: size mismatch for encoder.prompt_embeddings.weight: copying a param with shape torch.Size([480, 768]) from checkpoint, the shape in current model is torch.Size([2400, 768]).

I really appreciate any help you can provide. :)

[zhongwanjun]

Hi， To solve this issue, you need to modify the dimension of prompts from 100 to 20. In the code, it is set to 100 by default. Best, Wanjun 发件人: Soyeong Jeong ***@***.***> 日期: 星期四, 2023年3月9日 21:23 收件人: zhongwanjun/ProQA ***@***.***> 抄送: Subscribed ***@***.***> 主题: [zhongwanjun/ProQA] Runtime error when loading a pretrained model, ProQA (20 prompts+paq) (Issue #2) Hello, Thanks for sharing such an exciting work! However, I am facing an error when loading the pretrained model, ProQA (20 prompts+paq)<https://github.com/zhongwanjun/ProQA/releases/tag/model_release>. Specifically, I met a Runtime error when I run a zero-shot experiment, as follows: RuntimeError: Error(s) in loading state_dict for T5ForConditionalGeneration: size mismatch for encoder.prompt_embeddings.weight: copying a param with shape torch.Size([480, 768]) from checkpoint, the shape in current model is torch.Size([2400, 768]). I really appreciate any help you can provide. :) ― Reply to this email directly, view it on GitHub<#2>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AFNAK6NR2NNZ3VMFQKV2MTTW3HKUVANCNFSM6AAAAAAVVDTLQQ>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>