| [Doc] Add FSDP2 Muon optimizer feature guide
Co-authored-by: hanyyy<hanyue42@huawei.com>
# message auto-generated for no-merge-commit merge:
!2485 merge doc_fix into master
[Doc] Add FSDP2 Muon optimizer feature guide
Created-by: vasileone
Commit-by: hanyyy
Merged-by: ascend-robot
Description: ## What this PR does / why we need it?
新增 FSDP2 Muon 优化器特性文档,用于说明新 FSDP2 YAML 后端下如何启用和配置 Muon 优化器。文档包含:
- Muon 优化器的基本原理、优势与适用场景;
- 本仓库 Muon 实现与 FSDP2 DTensor 场景下的分片聚合与重新分片处理;
- 使用方法及optimizer: muon、matched_adamw_rms、muon_momentum、ns_steps 等配置项说明;
同时修正 FSDP2 开发者迁移文档中的重计算配置位置,将 recompute / recompute_plan.apply_modules 从 parallel 配置说明调整到 model 配置说明下,使其与当前 YAML 配置结构保持一致。
## Does this PR introduce any user-facing change?
Yes. Documentation-only change.
No code behavior or API changes are introduced.
## How was this patch tested?
Documentation-only patch.
Checked that the new Muon feature guide is under 300 lines and reviewed the updated Markdown content against the current YAML field names.
See merge request: Ascend/MindSpeed-MM!2485 | 19 天前 |