openai_text-embedding-ada-002/fixed_chunks/_chat_templating.txt_chunk_5.txt

Not all of the tool-calling features shown above are used by all models. Some use tool call IDs, others simply use the function name and
match tool calls to results using the ordering, and there are several models that use neither and only issue one tool 
call at a time to avoid confusion. If you want your code to be compatible across as many models as possible, we 
recommend structuring your tools calls like we've shown here, and returning tool results in the order that
they were issued by the model. The chat templates on each model should handle the rest.

Understanding tool schemas
Each function you pass to the tools argument of apply_chat_template is converted into a 
JSON schema. These schemas
are then passed to the model chat template. In other words, tool-use models do not see your functions directly, and they
never see the actual code inside them. What they care about is the function definitions and the arguments they
need to pass to them - they care about what the tools do and how to use them, not how they work! It is up to you
to read their outputs, detect if they have requested to use a tool, pass their arguments to the tool function, and
return the response in the chat.
Generating JSON schemas to pass to the template should be automatic and invisible as long as your functions
follow the specification above, but if you encounter problems, or you simply want more control over the conversion, 
you can handle the conversion manually. Here is an example of a manual schema conversion.
thon
from transformers.utils import get_json_schema
def multiply(a: float, b: float):
    """
    A function that multiplies two numbers
Args:
    a: The first number to multiply
    b: The second number to multiply
"""
return a * b

schema = get_json_schema(multiply)
print(schema)

This will yield:
json
{
  "type": "function", 
  "function": {
    "name": "multiply", 
    "description": "A function that multiplies two numbers", 
    "parameters": {
      "type": "object", 
      "properties": {
        "a": {
          "type": "number", 
          "description": "The first number to multiply"
        }, 
        "b": {
          "type": "number",
          "description": "The second number to multiply"
        }
      }, 
      "required": ["a", "b"]
    }
  }
}
If you wish, you can edit these schemas, or even write them from scratch yourself without using get_json_schema at 
all. JSON schemas can be passed directly to the tools argument of 
apply_chat_template - this gives you a lot of power to define precise schemas for more complex functions. Be careful,
though - the more complex your schemas, the more likely the model is to get confused when dealing with them! We 
recommend simple function signatures where possible, keeping arguments (and especially complex, nested arguments) 
to a minimum.
Here is an example of defining schemas by hand, and passing them directly to apply_chat_template:
thon
A simple function that takes no arguments
current_time = {
  "type": "function", 
  "function": {
    "name": "current_time",
    "description": "Get the current local time as a string.",
    "parameters": {
      'type': 'object',
      'properties': {}
    }
  }
}
A more complete function that takes two numerical arguments
multiply = {
  'type': 'function',
  'function': {
    'name': 'multiply',
    'description': 'A function that multiplies two numbers', 
    'parameters': {
      'type': 'object', 
      'properties': {
        'a': {
          'type': 'number',
          'description': 'The first number to multiply'
        }, 
        'b': {
          'type': 'number', 'description': 'The second number to multiply'
        }
      }, 
      'required': ['a', 'b']
    }
  }
}
model_input = tokenizer.apply_chat_template(
    messages,
    tools = [current_time, multiply]
)