deepspeedai / deepspeedai/DeepSpeed

[REQUEST] Can deepspeed initialize multiple datasets/dataloaders?

Open
#7,626 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement
Dominant language
Python
Stars
43.1k
Forks
5k
Avg merge
4d 15h
Merged PRs (30d)
112

Description

When I train a model, I want to evaluate the model and save ckpt at each epoch, enabling the same DP for the training dataloader and eval dataloader.

Now: DeepSpeed can only initialize the training dataset. If using the torch dataloader within the epoch, it will run duplicated.

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 by tracing DeepSpeed's current training-dataset initialization and how a torch DataLoader is consumed within an epoch. Determine the design needed to initialize separate training and evaluation dataloaders with the same data parallelism, without duplicated evaluation iteration. Done means epoch evaluation and checkpointing can use the separate loader correctly.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
distributed-systems, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.