support torchrun parameters and optimize gpu assignment

### 🚀 The feature

- support torchrun parameters in model config yaml file in frontend.
- optimize gpu assignment.

### Motivation, pitch

this is part of tasks in #2188 

### Alternatives

_No response_

### Additional context

_No response_