search

152334H / DL-Art-School

TorToiSe fine-tuning with DLAS
GNU Affero General Public License v3.0
205 stars 86 forks source link

Dataset path is not resolving in Windows #43

Closed tanfarou closed 1 year ago

tanfarou commented 1 year ago

The dataset path is not resolving; however, I can retrain on the older sets. Double checked the dataset, it has valid and train files and wavs folder.

Screenshot 2023-02-26 234242
152334H commented 1 year ago

This is not related to bad dataset path resolving (I think). This happens because of a recent PR (https://github.com/152334H/DL-Art-School/pull/40/files) to read files in UTF-8 (to account for different languages). This means that the dataset contains weird characters that are not in utf-8. The error should probably be made more ergonomic, but for the time being could you try to check what text encoding the valid.txt file is using?

tanfarou commented 1 year ago

This is not related to bad dataset path resolving (I think). This happens because of a recent PR (https://github.com/152334H/DL-Art-School/pull/40/files) to read files in UTF-8 (to account for different languages). This means that the dataset contains weird characters that are not in utf-8. The error should probably be made more ergonomic, but for the time being could you try to check what text encoding the valid.txt file is using?

Thank you! It is resolved. Valid.txt was in ANSI, and I changed it to utf-8, and it works perfectly.