Skip to content

Dataset #1

Description

@alecKarfonta

Anyone able to get some good output from this on another dataset?

I have tried to swap out the original dataset with the Cornell movie dialogue dataset but after training for two days on a big GPU I am getting garbage output. So now I am just trying to get it to output the input and again after two days of training the loss is very low but I am getting just a random collection of words. I have replaced the .en .fr files, and removed the old vocab files.

For another test I have created a dataset where the input is three letters of the alphabet and the output is the next three. This actually works, after a few minutes of training it gets perfect accuracy. Not sure why this won't translate to dialogue.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions