search

OpenMatch / TASTE

[CIKM 2023] This is the code repo for our CIKM‘23 paper "Text Matching Improves Sequential Recommendation by Reducing Popularity Biases".
MIT License
34 stars 4 forks source link

About evaluation during training #8

Closed tangzuoli closed 11 months ago

tangzuoli commented 12 months ago

The evaluation step is set to 5000 step in the model, while I follow your steps, I meet a problem in the validation during the training. I read your code and find that there is no prediction_step function in your trainer code which may result in a list error (a dictionary is need in the prediction_step while the inputs of your model is a list). I'm wondering should i set the eval_steps to 0 to make sure there is no validtion during the training and find the best ckeckpoint by the evaluate bash.

mssssss123 commented 11 months ago

Hi,

Based on your description, I think you may have the same problem like #3. This is a problem caused by the Transformer version being too high. Using versions below 4.24.0 should not have this problem.

Of course, it's also a workaround if you don't use evaluation during training.

Hope this helps you solve this problem!😊

tangzuoli commented 11 months ago

Thank you for your prompt response! I'm sorry that I didn't browse through the closed issues.