Gerald Woo comments

Results 55 comments of


                                            Gerald Woo

DDP (multi GPU) Iterable Dataset is not working as expected ?

Thanks a lot!

Bug when preparing data for finetuning

didn't look too deeply into this, but I'm guessing it's due to the format (column names) of your data frame? https://github.com/SalesforceAIResearch/uni2ts/blob/2ba614de8878d350c62835c942b450d2f4d5a711/src/uni2ts/data/builder/simple.py#L58

What causes train/PackedNLLLoss=-1.640?

https://github.com/SalesforceAIResearch/uni2ts/discussions/108#discussioncomment-10372349

Able to override concrete dunder methods

I didn't install from GitHub, here my versions of lineax and equinox: ``` >>> lx.__version__ '0.0.8' >>> eqx.__version__ '0.12.2' ``` I'm able to do this too: ```python class AbstractA(eqx.Module, strict=True):...

`seed_everything(..., workers=True)` causes the `Dataloader` to apply exactly the same augmentations each epoch if they sample values from `torch.distributions`

+1 seeing the same thing from seeds generated by `_generate_seed_sequence`. Base seeds yield random outputs, but after going through the `_generate_seed_sequence` function, the new seeds are equivalent to setting seed...