[BUG]'NotImplementedError: Cannot copy out of meta tensor; no data!' when loading T5/mistral from_config

[Yejing-Lai]

Describe the bug
When loading T5 and mistral from_config, The following error will appear：
'NotImplementedError: Cannot copy out of meta tensor; no data!'

To Reproduce
Steps to reproduce the behavior:

1. Simple inference script to reproduce
   reproduce script
2. What packages are required and their versions
   ipex: 2.1.100
   torch: 2.1
   oneccl 2.1.0
3. How to run the script
   deepspeed --num_gpus 2 --bind_cores_to_rank run_generation_with_deepspeed.py --benchmark -m mistralai/Mistral-7B-v0.1

Expected behavior
No error.

ds_report output
DeepSpeed C++/CUDA extension op report

NOTE: Ops not installed will be just-in-time (JIT) compiled at
runtime if needed. Op compatibility means that your system
meet the required dependencies to JIT install the op.

JIT compiled ops requires ninja
ninja .................. [OKAY]

op name ................ installed .. compatible

deepspeed_not_implemented [NO] ....... [OKAY]
deepspeed_ccl_comm ..... [NO] ....... [OKAY]
cpu_adam ............... [NO] ....... [OKAY]
fused_adam ............. [NO] ....... [OKAY]

DeepSpeed general environment info:
torch install path ............... ['xxx/lib/python3.9/site-packages/torch']
torch version .................... 2.1.2+cpu
deepspeed install path ........... ['xxx/deepspeed']
deepspeed info ................... 0.12.7+a85b6e47, a85b6e47, HEAD
deepspeed wheel compiled w. ...... torch 2.1

[Yejing-Lai]

PR can fix this issue.