Made changes in conda_env.yaml and train.py that solves issues related to decord. In the first one, decord was not a requirement and thus installation of an env was not complete without installing it. For the second one, importing decord before torch sometimes caused a segfault.
Made changes in
conda_env.yaml
andtrain.py
that solves issues related to decord. In the first one,decord
was not a requirement and thus installation of an env was not complete without installing it. For the second one, importingdecord
beforetorch
sometimes caused a segfault.This PR changes them both.