Izzy Putterman

To cover the bases, can you check how much disk space is free on your device?

Hey, Happy to hear that you are using the TSPP! We recommend using the NGC docker images which come with Apex installed. You can find how to build the TSPP...

I am not exactly sure what the C header issue is, but the recommended way of running the TSPP locally is to follow the Quick Start Guide: https://github.com/NVIDIA/DeepLearningExamples/tree/master/Tools/PyTorch/TimeSeriesPredictionPlatform#quick-start-guide. The Kaggle...

Hey, Not 100% sure in my answers here, but from my understanding, the Target columns can be calculated from the the Close column (https://www.kaggle.com/competitions/jpx-tokyo-stock-exchange-prediction/data). We decide to create the AdjustedClose...

Hey Yep, you are correct that it is caused by the addition of LazyModules specifically for the embedding. Please try to add these lines to here: https://github.com/NVIDIA/DeepLearningExamples/blob/master/Tools/PyTorch/TimeSeriesPredictionPlatform/training/trainer.py#L96 ```python dummy_batch, dummy_labels,...

Does your code run when using the old version of the TFT, I think currently present in the models/ directory? Can you train electricity with `TFT_SCRIPTING=false python launch_training.py model=tft dataset=electricity...

I am honestly not sure what the issue above is. Would you be able to re-preprocess the data and run again? What do you get if you use the old...

Awesome, glad that strangeness is fixed. I cannot give a timeline on release, but I will update this post when it is out.

@DaveBGld In what environment was the JPX data preprocessed? Did you copy the preprocessed data from the kaggle environment as well, or did you preprocess in the Docker?

This seems to be an error, that the element wise difference between the native and onnx models is above 1e-3. To get around this, you can try to comment out...