Antoine Honoré

periodic data export

12

I'd like to discuss a data export functionality. The main purpose I am seeing is the use of the data into other existing or home made co2 footprint calculation software....

Hi, I am looking for a way to use the code in a causal cross-attention setting, with an arbitrary causal mask that I could specify. e.g. ``` import torch from...

Write the model in a parallel framework which we do not need to maintain

1

I found this library: https://github.com/williamFalcon/pytorch-lightning As I understand it, they make it easy to parallelize torch models.

Single model multi GPU training

1

The discriminative step is now limited to the capacity of one GPU We must be able to perform the training of all single models on several GPUs.

Discriminative training was initially develop for a binary classification problem. Shortcuts might have been taken and prevent this to work for more than 2 classes.