Sajad Sotudeh

@patrickvonplaten To give you some clue about the potential root of the problem, given my experiment and [#](https://github.com/huggingface/transformers/issues/17106#issuecomment-1148125905), I believe this happens when the script deals with extremely large-scale datasets....

@n-hossain Hi, did you find the answer to this? I'm looking for the same _exact_ splits of data...

Hi @cylnlp I finally decided to go with the [Pegasus split](https://github.com/google-research/pegasus) of data; got the same response from the authors.

Hi @maxjeblick, Thank you for the reminder. Yes, we plan to release our best pre-trained model soon. I'll give you an update here, as soon as it's ready! Cheers, Sajad

Hi @apiro, Sorry for the slow response. I refactored the `data_builder.py` file. It works quite flawlessly now. If you've been using the proposed datasets, please download the latest version through...

Hi @CJPJ007, Thank you for reporting this! please see #3 Best, Sajad

@CJPJ007, what is your virtual machine's specification on Colab? I've been training (and testing) the model on my machine and it works quite fine. The specifications are as follows: NVIDIA...

Afaik, it doesn't, unless you come up with yet another architecture to do so (i.e., like adaptation). Stocks are multivariate, but TimesFM is for univariate!