How do I specify the chat format for the phi3.5 model? It seems like no one knows

[suhyun01150]

After setting chat_format to None, can I execute phi3, gemma, etc., exactly as described in the README?

llm.create_chat_completion(
messages = [
{"role": "system", "content": "You are an assistant who perfectly describes images."},
{
"role": "user",
"content": "Describe this image in detail please."
}
]
)

If it works, how does it do so? When I load the model, I see an output like this. Does it automatically recognize the format?
tokenizer.chat_template str = {% for message in messages %}{% if me...

somebody help me~~

[controlecidadao]

Hello!

I have been using it without specifying the prompt format and have obtained interesting results.
Check out our Samantha Interface Assistant project and take a look at the code:

https://github.com/controlecidadao/samantha_ia

This video shows Phi 3.5 vs Gemma 2 in an intelligence challenge about human nature using the interface. Llama 3.1 plays the role of judge:

https://www.youtube.com/watch?v=KgicCGMSygU