openclimatefix / openclimatefix/pvnet

Optional usage of DDP module in PyTorch

Open
#545 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

ocf-internal
Dominant language
Python
Stars
60
Forks
50
Avg merge
1d 21h
Merged PRs (30d)
2

Description

https://docs.pytorch.org/tutorials/intermediate/ddp_tutorial.html

For Raphael specifically, to try out DDP for multi-GPU training?

Thinking may be a good idea to have stated optionally, to speed up long training runs, or for when each step is too heavy for one GPU.

  • Can take over all the GPUs on single machine
  • Batch size is per GPU, so effective batch size increases with number of GPUs
  • Probably not much use as of the moment outside of cloucasting

Worth testing out perhaps and seeing what the speed up is. Should be pretty easy to try with Lightning, it's mostly just strategy="ddp" in the trainer.

Would definitely like a second opinion @dfulu

Contributor guide

No contributing guide indexed for this repository

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

Read the linked PyTorch DDP tutorial and identify the training entry point for Raphael. Evaluate whether Lightning's strategy="ddp" can be used for the stated multi-GPU cases, and measure the training-speed impact before documenting or proposing the optional usage.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.