meta-pytorch / meta-pytorch/data

online hard-example mining under Multi-GPU (DDP)

Open
#1,493 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
1.3k
Forks
179
Avg merge
6d 1h
Merged PRs (30d)
2

Description

🚀 The feature

Do you think it would be possible to have an off the shelf component like the one described at https://github.com/Lightning-AI/pytorch-lightning/discussions/1170?

Motivation, pitch

Sometime on unbalanced dataset it could be interesting to make specific sampling related to perplexity or othermetrics and loss values.
It could be also interesting for implementing some type of online curriculum learning/sampling solution.

Alternatives

No response

Additional context

Write and maintain a custom component.

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 with the linked PyTorch Lightning discussion and review this repository's data-loading utilities to identify an appropriate extension point. Clarify how sampling from perplexity or loss should work with Multi-GPU DDP, and define completion around an agreed component scope and distributed-training behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
data, distributed-systems, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.