deepspeedai / deepspeedai/DeepSpeed
Regarding lack of drop_last in Deepspeed Trainloader
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Hi ,
After encountering problems during training with my custom dataset using Deepspeed Initialize , I have narrowed the problem to the fact that since the number of images in my dataset is not divisible by the batch size , Deepspeed is throwing an error. This is easily resolvable in torch Dataloader by setting drop_last = True , the Deepspeed.pt.Dataloader source code however does not use this argument at all .
Could this be added to the source please ?
If not , please suggest an alternative means for training
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 DeepSpeed.pt.Dataloader source mentioned in the issue and compare its behavior with PyTorch DataLoader's drop_last argument. Reproduce training with a dataset whose size is not divisible by the batch size, then verify that the requested option prevents the reported error or that an alternative is documented.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning
- Issue type
- Feature
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100