Evaluate during training with distributed

maralm · February 21, 2020, 10:02pm

I am using the distributed training package to train on multiple gpus. Training works fine but I would like to be able to evaluate during training either on one gpu or multiple gpus. If I directly call evaluate function during training, each model produces different results. How can I get evaluation results every certain steps while using the distributed package for training?

ptrblck · February 22, 2020, 7:22am

If you are using DDP, the model replica should be initialized in the same manner. Since DDP performs an all-reduce step on gradients and assumes that they will be modified by the optimizer in all processes in the same way, the model output should be the same.
Are you also observing different outputs during training?

maralm · February 24, 2020, 5:57pm

When I evaluate during training, it runs on all gpus and each one produce different results. I am actually running this script:

github.com

huggingface/transformers/blob/master/examples/run_glue.py

# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa, Albert, XLM-RoBERTa)."""


import argparse
import glob

This file has been truncated. show original

On line 248, it is mentioned “Only evaluate when single GPU otherwise metrics may not average well”. I don’t understand why and how to change it to be able to evaluate correctly.