[Models] Add Phi3 to tests

[riedgar-ms]

Phi3 is now out, so start adding it into our testing frameworks

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 62.60%. Comparing base (acb38d1) to head (b29ad1e).
Report is 208 commits behind head on main.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #776      +/-   ##
==========================================
+ Coverage   55.37%   62.60%   +7.22%     
==========================================
  Files          55       55              
  Lines        4074     4070       -4     
==========================================
+ Hits         2256     2548     +292     
+ Misses       1818     1522     -296     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[BardiaKh]

Hi,
I am also interested in trying out Phi-3. I see that you used the default TransformerChat class, however that is not the chat template the model was trained with.
Any chance you could get the chat template to work?

Here is my code:

class Phi3Chat(TransformersChat, Transformers):
    """ A HuggingFace transformers version of the Phi-3 Chat language model with Guidance support.
    """
    def get_role_start(self, role_name, **kwargs):
        role_dict = {
            'user': "<|user|>\n",
            'assistant': "<|assistant|>\n",
            'system': "<|system|>\n"
        }
        return role_dict.get(role_name, '')
    
    def get_role_end(self, role_name=None):
        role_dict = {
            'user': "<|end|>\n",
            'assistant': "<|end|>\n",
            'system': "<|end|>\n"
        }
        return role_dict.get(role_name, '')

But this leads to tokenization error.