Arnaud Fickinger

Hi, I installed collider and coturn on a vm instance, it works well with the provided webversion test, but I got a POST error when I'm trying to use my...

Hi, thank you for open-sourcing the repo! Is there a particular reason why the exponential map of the Euclidean manifold is written as x+v/2 instead of x+v?

What is necessary in the training script?

2

I noticed that coordinator.block_all(), torch.set_num_threads(1) and dist.barrier() were added to the training script. Were they added for debugging purpose only or are they useful for training?

Gradients are None after booster.backward

10

After calling booster.backward(loss=loss, optimizer=optimizer), all gradients of model.module are None. Is there a way to access the gradients?

Could you please provide a snippet showing how to select the right layers (namely last 5-th layer and 7-th layer) as input to actionformer?