https://github.com/nixtla logo
t

Tuhin Mallick

04/10/2023, 3:11 PM
Hi 🙂, I have testing Temporal Fusion Transformer in the package, but I tend to run out of cuda memory.
Copy code
OutOfMemoryError: CUDA out of memory.
Is there a parameter to limit that or any recommendations on how to avoid that?
c

Cristian (Nixtla)

04/10/2023, 3:57 PM
Hi @Tuhin Mallick. In our experience TFT is highly memory intensive. You can limit the memory usage with
windows_batch_size
and
hidden_size
parameters
k

Kin Gtz. Olivares

04/10/2023, 6:01 PM
You can also try to reduce the size of the autorregresive features through the
input_size
Hey @Tuhin Mallick, We fixed this week TFT's memory usage: https://github.com/Nixtla/neuralforecast/pull/597 We will merge that PR into main soon. If you want to try the changes, you can install them with this line:
Copy code
!pip install git+<https://github.com/Nixtla/neuralforecast.git>