facebookresearch/fairseq

Python API Transformer.from_pretrained support directly to load on GPU

Open

#2,480 opened on Aug 13, 2020

View on GitHub
 (6 comments) (0 reactions) (0 assignees)Python (6,224 forks)batch import
enhancementhelp wanted

Repository metrics

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

Description

❓ Questions and Help

Hi,

I'm using the API command to load my finetuned mBART model, below is the command:

en2ar = TransformerModel.from_pretrained(
  '/home/ubuntu/mayub/models',
  checkpoint_file='checkpoint_best_en-ar_ft.pt',
  data_name_or_path='/home/ubuntu/mayub/models',
  bpe='sentencepiece',
  sentencepiece_vocab='sentencepiece.bpe.model'
)

Looks like this loads the model first on cpu and then I do en2ar.cuda() to replicate again on GPU device.

image

Is there a way I can avoid the loading on both devices and just directly load on GPU. Reason is I have 5 other models which I need to load and save on RAM.

Appreciate any help. Thanks !

Contributor guide

Research direction
Investigate the TransformerModel.from pretrained method and explore options to directly load the model on GPU, such as using the `device map` parameter or calling `.to('cuda')` within the loading process.
Tech stack
python
Domain
backend
Issue type
Research
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.
Under 1 hour
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
PythonPyTorch
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
70