app.py

import streamlit as st
import os
import requests

# From transformers import BertModel, BertTokenizer
from transformers import HfAgent, load_tool

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, Agent, LocalAgent

# checkpoint = "THUDM/agentlm-7b"
# model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map="auto", torch_dtype=torch.bfloat16)
# tokenizer = AutoTokenizer.from_pretrained(checkpoint)

# agent = LocalAgent(model, tokenizer)
# agent.run("Draw me a picture of rivers and lakes.")

# print(agent.run("Is the following `text` (in Spanish) positive or negative?", text="¡Este es un API muy agradable!"))

# Load tools
controlnet_transformer = load_tool("huggingface-tools/text-to-image")
upscaler = load_tool("diffusers/latent-upscaler-tool")


tools = [controlnet_transformer, upscaler]

# Define the custom HfAgent class
class CustomHfAgent(Agent):
    def __init__(
        self, url_endpoint, token=os.environ['HF_token'], chat_prompt_template=None, run_prompt_template=None, additional_tools=None
    ):
        super().__init__(
            chat_prompt_template=chat_prompt_template,
            run_prompt_template=run_prompt_template,
            additional_tools=additional_tools,
        )
        self.url_endpoint = url_endpoint
        self.token = token

    def generate_one(self, prompt, stop):
        headers = {"Authorization": self.token}
        inputs = {
            "inputs": prompt,
            "parameters": {"max_new_tokens": 192, "return_full_text": False, "stop": stop},
        }
        response = requests.post(self.url_endpoint, json=inputs, headers=headers)
        if response.status_code == 429:
            print("Getting rate-limited, waiting a tiny bit before trying again.")
            time.sleep(1)
            return self._generate_one(prompt)
        elif response.status_code != 200:
            raise ValueError(f"Errors {inputs} {response.status_code}: {response.json()}")

        result = response.json()[0]["generated_text"]
        # Inference API returns the stop sequence
        for stop_seq in stop:
            if result.endswith(stop_seq):
                return result[: -len(stop_seq)]
        return result

# Create the Streamlit app
st.title("Hugging Face Agent")

# Input field for the user's message
message_input = st.text_input("Enter your message:", "")

# Checkboxes for the tools to be used by the agent
tool_checkboxes = [st.checkbox(f"Use {tool}") for tool in tools]

# Submit button
submit_button = st.button("Submit")

# Define the callback function to handle the form submission
def handle_submission():
    # Get the user's message and the selected tools
    message = message_input.value
    selected_tools = [tool for tool, checkbox in tool_checkboxes]

    # Initialize the agent
    agent = CustomHfAgent(url_endpoint="https://api-inference.huggingface.co/models/bigcode/starcoder", token=os.environ['HF_token'])

    # Run the agent with the user's message and selected tools
    response = agent.run(message, tools=selected_tools)

    # Display the agent's response
    # Display the agent's response
    if response.startswith("Image:"):
        # Display the image response
        image_data = base64.b64decode(response.split(",")[1])
        img = Image.open(io.BytesIO(image_data))
        st.image(img)
    else:
        # Display the text response
        st.write(response)

# Add a button to trigger the agent to respond again
st.button("Ask Again")

# Define a callback function to handle the button click
def ask_again():
    # Reset the message input field
    message_input.value = ""

    # Run the agent again with an empty message
    agent.run("")

# Add the callback function to the button
st.button("Ask Again").do(ask_again)