add normalizedconfig for baichuan and mixformer-sequential

[changwangss]

What does this PR do?

Fixes # (issue)
we want to use the models https://huggingface.co/baichuan-inc/Baichuan2-13B-Base and https://huggingface.co/microsoft/phi-1, but the normalizedconfig doesn't contain.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[fxmarty]

Yes, normalizedconfig does not work well with custom models. WDYT @michaelbenayoun ?

[michaelbenayoun]

I think the issu is that we can end-up with multiple names potentially pointing to different NormalizedConfigs.

I think what we can do is allow the user to provide a NormalizedConfig on its own:

1. When exporting the model via the export function it is trivial to do so
2. When exporting the model via the CLI, it is less trivial: either the desired NormalizedConfig already exists and in this case it can simply be the name of the NormalizedConfig to use, or it does not exist yet, in which case I think we can point the user to use the export function?

[fxmarty]

couldn't we just be greedy and have a heuristic of most commonly used key names, and try to use those?

[michaelbenayoun]

Yes, we could do that.

[changwangss]

I think the issu is that we can end-up with multiple names potentially pointing to different NormalizedConfigs.

I think what we can do is allow the user to provide a NormalizedConfig on its own:

1. When exporting the model via the export function it is trivial to do so
2. When exporting the model via the CLI, it is less trivial: either the desired NormalizedConfig already exists and in this case it can simply be the name of the NormalizedConfig to use, or it does not exist yet, in which case I think we can point the user to use the export function?

Could you have the plan continue to do this feature? @fxmarty @michaelbenayoun
I close the PR first and hope the feature can be completed soon