Gorilla OpenFunctions

💡 Comes with Parallel Function Calling!
🚀 Try it out on Colab
📣 Read more in our OpenFunctions release blog

Introduction

Gorilla OpenFunctions extends Large Language Model(LLM) Chat Completion feature to formulate executable APIs call given natural language instructions and API context.

Models Available

Model	Functionality
gorilla-openfunctions-v0	Given a function, and user intent, returns properly formatted json with the right arguments
gorilla-openfunctions-v1	+ Parallel functions, and can choose between functions

All of our models are hosted on our Huggingface UC Berkeley gorilla-llm org: gorilla-openfunctions-v0 and gorilla-openfunctions-v1.

Example Usage

1. OpenFunctions is compatible with OpenAI Functions

!pip install openai==0.28.1

1. Point to Gorilla hosted servers

import openai

def get_gorilla_response(prompt="Call me an Uber ride type \"Plus\" in Berkeley at zipcode 94704 in 10 minutes", model="gorilla-openfunctions-v0", functions=[]):
  openai.api_key = "EMPTY"
  openai.api_base = "http://luigi.millennium.berkeley.edu:8000/v1"
  try:
    completion = openai.ChatCompletion.create(
      model="gorilla-openfunctions-v1",
      temperature=0.0,
      messages=[{"role": "user", "content": prompt}],
      functions=functions,
    )
    return completion.choices[0].message.content
  except Exception as e:
    print(e, model, prompt)

1. Pass the user argument and set of functions, Gorilla OpenFunctions returns a fully formatted json

query = "Call me an Uber ride type \"Plus\" in Berkeley at zipcode 94704 in 10 minutes"
functions = [
    {
        "name": "Uber Carpool",
        "api_name": "uber.ride",
        "description": "Find suitable ride for customers given the location, type of ride, and the amount of time the customer is willing to wait as parameters",
        "parameters":  [{"name": "loc", "description": "location of the starting place of the uber ride"}, {"name":"type", "enum": ["plus", "comfort", "black"], "description": "types of uber ride user is ordering"}, {"name": "time", "description": "the amount of time in minutes the customer is willing to wait"}]
    }
]
get_gorilla_response(query, functions=functions)

1. Expected output

uber.ride(loc="berkeley", type="plus", time=10)

Running OpenFunctions Locally

You can try this out on our Local OpenFunctions Colab to see how it works!

If you want to Run OpenFunctions locally, here is the prompt format that we used:

def get_prompt(user_query, functions=[]):
  if len(functions) == 0:
    return f"USER: <<question>> {user_query}\nASSISTANT: "
  functions_string = json.dumps(functions)
  return f"USER: <<question>> {user_query} <<function>> {functions_string}\nASSISTANT: "

Note: Use the get_prompt format only if you are hosting it Locally. If you are using the Berkeley hosted models through the Chat-completion API, we do this in the backend, so you don't have to do this. The model is supported in Hugging Face 🤗 Transformers and can be run up locally:

import json
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

def get_prompt(user_query: str, functions: list = []) -> str:
    """
    Generates a conversation prompt based on the user's query and a list of functions.

    Parameters:
    - user_query (str): The user's query.
    - functions (list): A list of functions to include in the prompt.

    Returns:
    - str: The formatted conversation prompt.
    """
    if len(functions) == 0:
        return f"USER: <<question>> {user_query}\nASSISTANT: "
    functions_string = json.dumps(functions)
    return f"USER: <<question>> {user_query} <<function>> {functions_string}\nASSISTANT: "

# Device setup
device : str = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Model and tokenizer setup
model_id : str = "gorilla-llm/gorilla-openfunctions-v0"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True)

# Move model to device
model.to(device)

# Pipeline setup
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=128,
    batch_size=16,
    torch_dtype=torch_dtype,
    device=device,
)

# Example usage
query: str = "Call me an Uber ride type \"Plus\" in Berkeley at zipcode 94704 in 10 minutes"
functions = [
    {
        "name": "Uber Carpool",
        "api_name": "uber.ride",
        "description": "Find suitable ride for customers given the location, type of ride, and the amount of time the customer is willing to wait as parameters",
        "parameters":  [
            {"name": "loc", "description": "Location of the starting place of the Uber ride"},
            {"name": "type", "enum": ["plus", "comfort", "black"], "description": "Types of Uber ride user is ordering"},
            {"name": "time", "description": "The amount of time in minutes the customer is willing to wait"}
        ]
    }
]

# Generate prompt and obtain model output
prompt = get_prompt(query, functions=functions)
output = pipe(prompt)

print(output)

Self-Hosting OpenFunctions

This section provides a guide on how to self-host the OpenFunctions model on your local machine or serve it locally for your enterprise. The server deploys the OpenFunctions-v0 model with uvicorn, while the client interacts with this local server using the OpenAI package (0.28.xx).

Setting Up Your Local Server

The server API endpoint mirrors the interface of the API call executed by openai.ChatCompletion.create, ensuring compatibility with clients using the OpenAI package.

Ensure you have the required libraries:

pip install fastapi uvicorn transformers torch

Similar to the process outlined in the "Running OpenFunctions Locally" section, the model is loaded from HuggingFace for local serving. Only use the get_prompt format if you are hosting locally.

from fastapi import FastAPI
from pydantic import BaseModel
import torch
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
import uvicorn
import json

# Define the API request model
class ChatCompletionRequest(BaseModel):
    model: str
    temperature: float
    messages: list
    functions: list = []

# Initialize the FastAPI app
app = FastAPI()

# Device setup
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Model and tokenizer setup
model_id = "gorilla-llm/gorilla-openfunctions-v0"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True)

# Move model to device
model.to(device)

# Pipeline setup
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=128,
    batch_size=16,
    torch_dtype=torch_dtype,
    device=device,
)

# Function to generate prompt
def get_prompt(user_query: str, functions: list) -> str:
    """
    Generates a conversation prompt based on the user's query and a list of functions.

    Parameters:
    - user_query (str): The user's query.
    - functions (list): A list of functions to include in the prompt.

    Returns:
    - str: The formatted conversation prompt.
    """
    if len(functions) == 0:
        return f"USER: <<question>> {user_query}\nASSISTANT: "
    functions_string = json.dumps(functions)
    return f"USER: <<question>> {user_query} <<function>> {functions_string}\nASSISTANT: "

# API endpoint
@app.post("/chat/completions")
async def chat_completion(request: ChatCompletionRequest):
    user_query = request.messages[0]['content']
    prompt = get_prompt(user_query, request.functions)
    output = pipe(prompt)
    generated_text = output[0]['generated_text']

    # Return in the format expected by OpenAI package
    return {
        "id": "0",
        "object": "text_completion",
        "created": 0,
        "model": request.model,
        "choices": [{"text": generated_text, "index": 0, "logprobs": None, "finish_reason": "length"}]
    }

# Run the server
if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)

Setting Up the Client

Ensure you have the required libraries:

!pip install openai==0.28.1

The example client below demonstrates how to interact with the locally hosted OpenFunctions model using openai.ChatCompletion.create, akin to using Gorilla hosted servers.

import openai

def get_gorilla_response(prompt="Call me an Uber ride type \"Plus\" in Berkeley at zipcode 94704 in 10 minutes", model="gorilla-openfunctions-v0", functions=[]):
    """
    Sends a request to the self-hosted OpenFunctions model and retrieves the response.

    This function configures the OpenAI API to interact with a local server instance of the OpenFunctions model. 
    It sends a prompt along with optional model specifications and functions, then returns the model's text response.

    Parameters:
    - prompt (str): The user's query.
    - model (str): The identifier for the specific model version to be used. Defaults to "gorilla-openfunctions-v0".
    - functions (list): An optional list of functions providing context for the model's response.

    Returns:
    - str: The text response generated by the model.

    Raises:
    - Exception: If there's an issue with processing the request or communicating with the server.
    """
    openai.api_key = "EMPTY"
    openai.api_base = "http://localhost:8000"  # Point to the local server
    try:
        completion = openai.ChatCompletion.create(
            model="gorilla-openfunctions-v0",
            temperature=0.0,
            messages=[{"role": "user", "content": prompt}],
            functions=functions,
        )
        return completion.choices[0].text
    except Exception as e:
        print(e, model, prompt)

# Example usage
query: str = "Get the latest news headlines from CNN."
functions = [
    {
        "name": "News Headlines",
        "api_call": "news.get_headlines",
        "description": "Retrieve the latest news headlines from a specific news source.",
        "parameters": {
            "type": "object",
            "properties": {
                "source": {
                    "type": "string",
                    "description": "The news source, e.g. CNN"
                }
            },
            "required": [
                "source"
            ]
        }
    }
]

resp = get_gorilla_response(query, functions=functions)
print(resp)

Try Out Self-Hosting OpenFunctions on Colab

You can try out setting up a local server to self-host the OpenFunctions model using this OpenFunctions Self-Hosted Colab notebook. Make sure you select an A100/V100 instance to run the notebook – smaller instances like T-4 do not suffice due to memory constraints. To remotely access the server running on the Colab instance from a local client, ngrok is used to tunnel the server ports from the Colab instance to public URLs. Instructions for setting up ngrok are provided in the notebook.

Evaluation

Contributing

All the models, and data used to train the models is released under Apache 2.0. Gorilla is an open source effort from UC Berkeley and we welcome contributors. Please email us your comments, criticism, and questions. More information about the project can be found at https://gorilla.cs.berkeley.edu/