Lightning-AI / Lightning-AI/pytorch-lightning

Before/After dataloader batch callback hooks

Open
#13,095 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

callback question won't fix
Dominant language
Python
Stars
31.4k
Forks
3.8k
Avg merge
6d 7h
Merged PRs (30d)
6

Description

## 🚀 Feature

I would like to have callback hooks (or module hooks) before and after we try to get the next batch from the torch dataloader, in order to know how much time is spent waiting for data to be loaded

### Motivation

I am frustrated with the lack of any ability to know how long it takes for a batch to be ready from the torch dataloader. In a pytorch script, I could simply time it takes in the training loop:
```
start = time.time()
for batch in dataloader:
time_taken = time.time() - start
start = time.time()
```

### Pitch

Callback hooks seem to be the right solution here.
### Alternatives
I subclassed the pytorch Dataloader object and replaced the iterator it returns decorated with a spooky function that replaces the __next__ call with a new function that returns __next__ while also timing.

### Additional context
If there is some way to do this that I am not seeing, please let me know
______________________________________________________________________

cc @awaelchli @ananthsub @rohitgr7

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

No project files or tests are named. Start by locating the training-loop integration around PyTorch DataLoader iteration and compare it with the timing example in the issue. Done means users can observe hooks immediately before and after waiting for the next batch, with the behavior covered by tests.

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
Stale
Clarity
Mostly clear
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.