pytorch / pytorch/vision

Multi GPU support for vision model training in C++ with libtorch

Open
#3,052 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

module: c++ frontend
Dominant language
Python
Stars
17.9k
Forks
7.3k
Avg merge
1d 15h
Merged PRs (30d)
13

Description

🚀 Feature

Multi GPU support for vision model training in C++ with libtorch.

Motivation

More and more systems rely only on C++ for deep learning, including ours. But at the moment the C++ version of vision models are quite unpractical to use, and it is not possible to train them on multiple GPU when using large amount of data.

Pitch

Make C++ models support multi GPU. This can be done by deriving the pytorch ICloneable class, but this might not be sufficient for optimal GPU training, as pointed out in this issue. The best option might be to implement the clone method so that data_parallel (here) can handle the models.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing the C++ vision model interfaces and the referenced torch/csrc/api/include/torch/nn/parallel/data_parallel.h entry point, along with the linked PyTorch issue about ICloneable. Determine how clone and data_parallel are expected to interact for these models. Done means C++ vision models can train across multiple GPUs with large datasets.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, pytorch
Domain
computer-vision, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.