[Docs] Setting JSON mode on the "OpenGVLab/InternVL2-40B" model.

### 📚 The doc issue

Hello all,

On the HuggingFace demo, there is a mention that the Hermes2 model can be used as the ChatTemplateConfig for the "OpenGVLab/InternVL2-40B" model like below:
`chat_template_config = ChatTemplateConfig('internvl-zh-hermes2')`

This model supports JSON Mode, whereby outputs can be specified to be in JSON format. It is described [here](https://huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B/blob/main/README.md)
When using offline inference using the OpenAI prompt style, how can I enable or make use of this?

Are there other ways that JSON mode can be enforced using this model?

Thank you for your assistance and development of this repo!

### Suggest a potential alternative/fix

_No response_