mode = instructor.Mode.TOOLS vs. instructor.Mode.PARALLEL_TOOLS

[Mr-Ruben]

Forgive me if it is in the documentation (couldn't find it), but what is the point of having 2 ways to instantiate the client

# Normal Tools
client = instructor.from_openai(openai.OpenAI(),
                               mode=instructor.Mode.TOOLS,
                               )

# Parallel Tools
client = instructor.from_openai(openai.OpenAI(),
                               mode=instructor.Mode.PARALLEL_TOOLS,
                               )

if you can obtain a full_response AND multiple calls (= tools) just by making the response_model class an Iterable?

class UserExtract(BaseModel):
    name: str
    age: int

class MovieExtract(BaseModel):
    movie: str
    year: int

# Create a class that has an Iterable of classes (to allow Multiple Tools)
class Output(BaseModel):
   #results: Union[UserExtract, MovieExtract]   # Before, just one OR the other
   results: Iterable[Union[UserExtract, MovieExtract]]  # Now, any number

With this you obtain full response + multiple tools

# Call API
output, assistant_message_response = client.chat.completions.create_with_completion(
    model="gpt-3.5-turbo",
    response_model=Output,
    messages=[
        {"role": "user", "content": "Jason is a guy of 25 years old. Today he is watching `Titanic (2002)`"},
    ],
)

# Display
for call in output.results:
    call

# UserExtract(name='Jason', age=25)
# MovieExtract(movie='Titanic', year=2002)