fix system prompts

[Cmxu]

No description provided.

[Cmxu]

Need to give 'type': 'text' for content. Current implementation causes system prompt to be ignored. For example, with LLaVA-Next-Video:

prompt = [{'role': 'system', 'content': [{'text': 'Be Helpful.'}]}, {'role': 'user', 'content': [{'type': 'text', 'text': 'Help.'}, {'type': 'video'}]}]
processor.apply_chat_template(prompt, add_generation_prompt = True)

Results in
USER: <video>\nHelp. ASSISTANT:

But,

prompt = [{'role': 'system', 'content': [{'type': 'text', 'text': 'Be Helpful.'}]}, {'role': 'user', 'content': [{'type': 'text', 'text': 'Help.'}, {'type': 'video'}]}]
processor.apply_chat_template(prompt, add_generation_prompt = True)

Results in
Be Helpful. USER: <video>\nHelp. ASSISTANT:

[zjysteven]

Thank you for the PR and sorry for my late response. This is a good catch. I will investigate a bit before merging it though, because I remember clearly that for some models (e.g., LLaVA-1.5) the system prompts work fine.