mindspore-ai / mindspore-ai/hyper-parallel
[Bug]在fully_shard接口传入replicate_params时，第一个step grad_norm异常比baseline大。
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 53
- Forks
- 63
- Avg merge
- 23h 45m
- Merged PRs (30d)
- 63
Description
该问题是怎么引起的？
网络中部分参数采取不切，将这部分参数传给replicate_params，拉起训练，第一个step grad nrom较大。其他参数配置不改
重现步骤
报错信息
不报错，grad norm异常，baseline grad norm是28.97633362。传replicate_params参数时训练的grad norm是48.47068405
schema_version: 1
source: gitcode
gitcode_repo: mindspore/hyper-parallel
gitcode_issue: 129
source_url: https://gitcode.com/mindspore/hyper-parallel/issues/129
Contributor guide
No contributing guide indexed for this repository
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 at the fully_shard interface and reproduce the first training step with and without replicate_params, keeping all other parameters unchanged. Compare the resulting grad norm with the reported baselines of 28.97633362 and 48.47068405, then trace the discrepancy and verify that the first-step value matches the baseline after the fix.
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
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100