facebookresearch/fairseq

Why inplace operation in `AdaptiveSoftmax`

Open

#760 opened on May 28, 2019

View on GitHub
 (6 comments) (1 reaction) (1 assignee)Python (6,224 forks)batch import
enhancementhelp wantedstale

Repository metrics

Stars
 (29,107 stars)
PR merge metrics
 (No merged PRs in 30d)

Description

I used log_output for training purpose, but pytorch complains about inplace operation on backward.

For lines 204 and 205: https://github.com/pytorch/fairseq/blob/e6422528dae0b899848469efe2dc404c1e639ce9/fairseq/modules/adaptive_softmax.py#L200-L208

  1. why doing a copy_ on tailout
  2. why inplace add_

I changed to

tail_output = ...
... = self.lsm(tail_out) + tail_priors[idxs, i, None]

And it works for training.

Contributor guide

Research direction
Examine the inplace operations in adaptive softmax.py and replace them with out of place versions to allow gradient computation.
Tech stack
pythonpytorch
Domain
backend
Issue type
Bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
2
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
PyTorch
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60