Skip to content

Instantly share code, notes, and snippets.

@thomwolf
thomwolf / parallel.py
Last active August 8, 2023 15:50
Data Parallelism in PyTorch for modules and losses
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
## Created by: Hang Zhang, Rutgers University, Email: zhang.hang@rutgers.edu
## Modified by Thomas Wolf, HuggingFace Inc., Email: thomas@huggingface.co
## Copyright (c) 2017-2018
##
## This source code is licensed under the MIT-style license found in the
## LICENSE file in the root directory of this source tree
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
"""Encoding Data Parallel"""