deepspeedai / deepspeedai/DeepSpeed
[REQUEST] multiple dynamic loss scalar on different loss values
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Is your feature request related to a problem? Please describe.
We have different losses in each iteration, and we'd like to have different dynamic loss scalar to stable the iteration. However, the current deepspeed only supports one scalar if I understand correctly.
Describe the solution you'd like
I'm not sure how deepspeed can implement this feature, but a simple apex implementation can be as follows by specifying different loss id.
with amp.scale_loss(losses, self.optimizer, loss_id=loss_id) as scaled_loss:
scaled_loss.backward()
Describe alternatives you've considered
Currently, for stability, we used apex, which does not have the feature of ZERO.
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
The issue names no DeepSpeed files or tests. Start by tracing the dynamic loss-scaling path around the optimizer and compare it with the shown Apex amp.scale_loss(..., loss_id=...) example. Done means different losses can use different scalars while retaining DeepSpeed's ZERO support.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning, performance
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100