Added --chat-template-file to llama-run

[engelmi]

Relates to: #11178

Added --chat-template-file CLI option to llama-run. If specified, the file will be read and the content passed for overwriting the chat template of the model to common_chat_templates_from_model.

This also enables running the granite-code model from ollama:

# using a jinja chat template file 
# (when prefix, e.g. hf://, is not specified, llama-run pulls from ollama)
$ llama-run  --chat-template-file ./chat.tmpl granite-code
> write code

Here is a code snippet in Python:

"""
def f(x):
    return x**2
"""

# without a jinja chat template file
$ llama-run granite-code
> write code
failed to apply the chat template

Make sure to read the contributing guidelines before submitting a PR

[engelmi]

Preceding PR: #11922

And still this error even though there is no merge commit:

Merge is not an allowed merge method in this repository.
This branch must not contain merge commits. 

[ericcurtin]

We should do:

chat_template.empty() ? "" : chat_template

here. Passing nullptr to a reference is not allowed. I wish the compiler caught these things.

common_chat_templates_ptr common_chat_templates_init(
                                    const struct llama_model * model,
                                           const std::string & chat_template_override,
                                           const std::string & bos_token_override = "",
                                           const std::string & eos_token_override = "")

[ericcurtin]

Maybe the std::string class is smart enough to interpret all these as the same thing:

"", '', 0, NULL, nullptr

and that's why it compiles/works 🤷 . So it might be just implicitly converting it to "".