【Atlas800型号9000】【新需求】【模型训练】mindspore去ranktable方案大模型通信要占用3-4G，占用更多device内存

## Environment
- **Hardware Environment(`Ascend`/`GPU`/`CPU`)**: 
Ascend Data Center Solution 23.0.RC1.B100
驱动和固件:Ascend HDK 23.0.RC1.B100
https://cmc-szv.clouddragon.huawei.com/cmcversion/index/releaseView?deltaId=8124113787289600&isSelect=Software
> Uncomment only one ` /device <>` line, hit enter to put that in a new line, and remove leading whitespaces from that line:
> /device ascend

- **Software Environment**:
-- MindSpore version (source or binary):Mindspore2.0.0.B230
-- Python version (e.g., Python 3.7.5):Python 3.7.5
-- OS platform and distribution (e.g., Linux Ubuntu 16.04):4.19.36-vhulk1907.1.0.h475.eulerosv2r8.aarch64
-- GCC/Compiler version (if compiled from source): 
https://cmc-szv.clouddragon.huawei.com/cmcversion/index/releaseView?deltaId=5276011499749760&isSelect=Software
## Related testcase
PanGu 13B训练
## Steps to reproduce the issue
1. Pangu 13B占用内存3-4g

## Describe the current behavior
内存正常。
## Describe the expected behavior
每个通信group要申请200M左右的device内存，大模型要3-4G

## Related log / screenshot

## Special notes for this issue
每个通信group要申请200M左右的device内存，大模型要3-4G，175B可能影响更大
## 定位开发：周培晨 00425770

MindSpore/models

内容风险标识

评论 (2)

MindSpore/models .gitee-modal { width: 500px !important; }

内容风险标识