[Bug]: 初始化阶段会有两次权重加载过程

Checklist

• 1. I have searched the existing issues (https://gitcode.com/mindspore/hyper-parallel/issues)
• 2. I have read the relevant documentation.
• 3. I have created a minimal reproduction case that clearly demonstrates the issue, including a complete code example and the error message with full traceback and error logs.

🐛 Describe the bug

使用Qwen3模型进行训练，在初始化阶段会有两次权重加载过程：

1. 第一次是在meta初始化阶段from_pretrained会触发读权重到cpu，大概耗时2~5min，有进度条提醒
2. 第二次是分片后实际初始化阶段，耗时反而更久10min左右，且没加载进度打印

Expected behavior

首次只做meta初始化获取到model，分片后再加载权重

Additional context

Environment info

Python 3.11.10
torch 2.6.0
transformers 5.13.0

Thanks for contributing 🎉!

schema_version: 1
source: gitcode
gitcode_repo: mindspore/hyper-parallel
gitcode_issue: 365
source_url: https://gitcode.com/mindspore/hyper-parallel/issues/365

Start with the Qwen3 training reproduction and trace the initialization path around from_pretrained, comparing the meta-initialization and post-sharding stages. Confirm which stage reads weights and where loading progress is reported. Done means initialization performs only the intended weight load after sharding, without the earlier duplicate load.