liaoyuhua / tempo-pytorch Goto Github PK

So is TEMPO really effective?

Hello,

I've been encountering difficulties in getting my model to run on Kaggle. Could you kindly share how you managed to successfully execute your model there?
https://www.kaggle.com/code/mrantonzaitsev/tempo?scriptVersionId=165970347
Thank you for your help!

Hello @liaoyuhua
Thanks for making the tempo-pytorch implementation available
But I am facing a few issues while running it on a GPU cluster (Out of Memory issue), which says
CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 15.78 GiB total capacity; 13.53 GiB already allocated; 7.75 MiB free; 13.55 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Can you please provide some information on how were you able to run this (compute details if possible) and the time taken for you to carry out the training process?

Thanks in advance for your help

Thank you very much for your implementation!!! but why is the result (Test MAE) I got very large, like 7.42 ? I didn't modify the code, is it because I missed something?

Hello.
How to train the model on a different sample size and make a forecast for a different forecast horizon
I can't do it.

Hi @liaoyuhua,

First of all, I would like to thank you for sharing the code for your Tempo-pytorch implementation. It is a valuable resource for the community and I appreciate your efforts in making it available.

I have a few questions about the Prompt Pool implementation in your code. I am particularly interested in:

How are the prompts in the pool initialized?
How is the prompt pool updated over time?

作者你好，
复现的工作很出色，但是prompt pool的实现方法中并没有计算key的loss，与 https://github.com/JH-LEE-KR/l2p-pytorch/blob/main/prompt.py 中官方实现有一定差别。