Add keep_torch_compile param to unwrap_model and extract_model_from_parallel for distributed compiled model.
#5239
| Job | Run time |
|---|---|
| 2m 47s | |
| 2m 47s |