app.py

import gradio as gr
from huggingface_hub import InferenceClient, list_models
import os
import json

# Initialize Hugging Face Inference Client
api_key = os.getenv("HF_TOKEN")
client = InferenceClient(api_key=api_key)

# Load or initialize system prompts
PROMPTS_FILE = "system_prompts.json"
if os.path.exists(PROMPTS_FILE):
    with open(PROMPTS_FILE, "r") as file:
        system_prompts = json.load(file)
else:
    system_prompts = {"default": "You are a good image generation prompt engineer for diffuser image generation models"}

def save_prompts():
    """Save the current system prompts to a JSON file."""
    with open(PROMPTS_FILE, "w") as file:
        json.dump(system_prompts, file, indent=4)

def chat_with_model(user_input, system_prompt, selected_model):
    """Send user input to the model and return its response."""
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": user_input}
    ]

    try:
        result = client.chat.completions.create(
            model=selected_model,
            messages=messages,
            temperature=0.5,
            max_tokens=2048,
            top_p=0.7,
            stream=False  # Stream disabled for simplicity
        )
        return result["choices"][0]["message"]["content"]
    except Exception as e:
        return f"Error: {str(e)}"

def update_prompt(name, content):
    """Update or add a new system prompt."""
    system_prompts[name] = content
    save_prompts()
    return f"System prompt '{name}' saved."

def get_prompt(name):
    """Retrieve a system prompt by name."""
    return system_prompts.get(name, "")

    
def update_dropdown_choices(models):
    """Update dropdown choices dynamically."""
    if not models:
        return [], None  # Return empty choices and no default value
    return models, models[0]



def fetch_models(task):
    """Fetch models for a specific task from Hugging Face Hub."""
    try:
        models = list_models(filter=f"pipeline_tags:{task}")
        models += [
            "HuggingFaceH4/zephyr-7b-beta",
            "HuggingFaceH4/zephyr-7b-alpha",
            "HuggingFaceH4/zephyr-6b"
        ]

        return [model.modelId for model in models]
    except Exception as e:
        return [f"Error fetching models: {str(e)}"]
task_choices=["text-generation", "image-classification", "text-classification", "translation"]
# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## Hugging Face Chatbot with Dynamic Model Selection")

    with gr.Row():
        with gr.Column():
            # Task selection
            task_selector = gr.Dropdown(
                choices=task_choices,
                label="Select Task",
                value="text-generation"
            )
            models=fetch_models(task_choices[0])
            # Model selector
            model_selector = gr.Dropdown(choices=[], label="Select Model")
            # Example of handling model updates
            model_selector.choices = models
            model_selector.value = models[0]

            # System prompt and input
            system_prompt_name = gr.Dropdown(choices=list(system_prompts.keys()), label="Select System Prompt")
            system_prompt_content = gr.TextArea(label="System Prompt", value=get_prompt("default"), lines=4)
            save_prompt_button = gr.Button("Save System Prompt")

            user_input = gr.TextArea(label="Enter your prompt", placeholder="Describe the character or request a detailed description...", lines=4)
            submit_button = gr.Button("Generate")

        with gr.Column():
            output = gr.TextArea(label="Model Response", interactive=False, lines=10)

    # Update model list when task changes
    def update_model_list(task):
        models = fetch_models(task)
        return gr.Dropdown.update(choices=models, value=models[0] if models else None)

    # Event bindings
    task_selector.change(update_model_list, inputs=[task_selector], outputs=[model_selector])
    save_prompt_button.click(update_prompt, inputs=[system_prompt_name, system_prompt_content], outputs=[])
    submit_button.click(chat_with_model, inputs=[user_input, system_prompt_content, model_selector], outputs=[output])

# Run the app
demo.launch()