mindspore-ai / mindspore-ai/hyper-parallel
支持非融合 FSDP 逐参数混合 dtype
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 53
- Forks
- 63
- Avg merge
- 23h 45m
- Merged PRs (30d)
- 63
Description
当前 MindSpore FSDP 非融合逐参数通信路径仍复用 state 级统一 dtype 元数据，混合原始参数 dtype 时会触发统一 dtype 约束或导致错误 cast。需要改为按参数自身的 orig_dtype/reduce_dtype 进行梯度通信与回写，并保留 comm_fusion 路径的统一 dtype 约束。
schema_version: 1
source: gitcode
gitcode_repo: mindspore/hyper-parallel
gitcode_issue: 285
source_url: https://gitcode.com/mindspore/hyper-parallel/issues/285
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 by locating the non-fused, per-parameter FSDP gradient communication and write-back path, then trace where state-level dtype metadata is applied. Check mixed original parameter dtypes and verify that each parameter uses its own orig_dtype/reduce_dtype, while the comm_fusion path retains its uniform dtype constraint.
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
- Mostly clear
- Newbie friendliness
- 52/100