Support molmo in turbomind

[lvhan028]

model: allenai/Molmo-7B-D-0924

[irexyc]

It seems doesn't support text only input like

pipe('who are you?')

# or
messages = [
    dict(role='user', content=[
        dict(type='text', text='你是谁')
    ])
]
pipe(messages)

[irexyc]

When adding elements to out_messages, parameter i should be considered

[irexyc]

And after fixing the order, I find the position of Assistant seems not right. I decode the input token and the result is

" User: Describe the two images in detail. Assistant:Assistant: xxx...User: What are the similarities and differences between these two images. "  

(use the multi-round example here)

If the input only has one user content and decode the input ids, the result is

" User: Describe the two images in detail. Assistant:"

[lvhan028]

Fixed

[lvhan028]

It seems doesn't support text only input like

pipe('who are you?')

# or
messages = [
    dict(role='user', content=[
        dict(type='text', text='你是谁')
    ])
]
pipe(messages)

Fixed