open-mmlab / open-mmlab/mmengine
[Feature] Make advance dataloader optional in iterbased training loop
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.5k
- Forks
- 464
- PR merge metrics
- No merged PRs in 30d
Description
What is the feature?
Sometimes is not necessary to iterate the dataloader when resuming, mostly when the training batch is randomly sampled with replacement. Also, When working in the cloud, GPUs are reserved without being actually used during the advance dataloader iteration, which jeopardizes the management of resources.
It would be nice to make this step optional. Can we add an argument to the cfg options to skip this step?
Hope this request is useful.
Any other context?
No response
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by locating the iterbased training loop and the resume path that advances the dataloader. Trace how configuration options are defined and consumed, then verify that the advance step can be skipped for selected resumes while existing resume behavior remains unchanged.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100