modelscope / modelscope/ms-swift
DeepSpeed with Muon Optimizer
Open
Nobody has claimed this yet.
stale
- Dominant language
- Python
- Stars
- 15.7k
- Forks
- 1.7k
- Avg merge
- 1d 16h
- Merged PRs (30d)
- 136
Description
It seems that using muon alone (zero0) is also not working. I tested deespeed zero1, 2, all lead to the same error. Please consider supporting Muon Optimizer alongside deepseed.
swift sft
--deepspeed zero0 \
--optimizer muon \
...
[rank0]: File "/root/.cache/modelscope/Moonlight/examples/toy_train.py", line 60, in zeropower_via_newtonschulz5
[rank0]: assert len(G.shape) == 2
[rank0]: ^^^^^^^^^^^^^^^^^
[rank0]: AssertionError
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
Reproduce the failure with the shown swift sft command using --deepspeed zero0 and --optimizer muon, then inspect examples/toy_train.py at zeropower_via_newtonschulz5 line 60. Trace how optimizer parameters are passed under the DeepSpeed configurations and verify completion without the len(G.shape) == 2 assertion.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100