import gradio as gr
import os
from huggingface_hub import InferenceClient
from huggingface_hub.inference._generated.types.chat_completion import ChatCompletionStreamOutput

# Use the fine-tuned maritime legal model
MODEL = "nomiChroma3.1"
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

from huggingface_hub.inference._generated.types.chat_completion import ChatCompletionStreamOutput

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    messages = [{"role": "system", "content": system_message}]
    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})
    messages.append({"role": "user", "content": message})
    response = ""
    try:
        for message in client.chat_completion(
            messages,
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
        ):
            try:
                if isinstance(message, ChatCompletionStreamOutput):
                    content = message.choices[0].delta.content
                    if content is not None:
                        response += content
                        yield response
                    if message.choices[0].finish_reason == 'stop':
                        break
                elif isinstance(message, dict):
                    content = message.get('choices', [{}])[0].get('delta', {}).get('content')
                    if content:
                        response += content
                        yield response
                    if message.get('choices', [{}])[0].get('finish_reason') == 'stop':
                        break
                elif isinstance(message, str):
                    if message.strip():  # Only process non-empty strings
                        response += message
                        yield response
                else:
                    print(f"Unexpected message type: {type(message)}")
                    print(f"Message content: {message}")
            except Exception as e:
                print(f"Error processing message: {e}")
                print(f"Problematic message: {message}")
                continue  # Continue to the next message even if there's an error
        
        # Final yield to ensure all content is returned
        if response:
            yield response

    except Exception as e:
        print(f"An error occurred in the main loop: {e}")
        if response:
            yield response
        else:
            yield f"An error occurred: {e}"

# Gradio interface setup
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(
            value="You are a maritime legal assistant with expertise strictly in Indian maritime law. Provide detailed legal advice and information based on Indian maritime legal principles and regulations.",
            label="System message"
        ),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
    title="Maritime Legal Compliance",
    description="This chatbot uses Fine-tuned LLAMA-3.1 model personalised specifically to provide assistance with Indian maritime legal queries.",
    theme="soft",
    examples=[
        ["What are the key regulations governing ports in India?"],
        ["Explain the concept of cabotage in Indian maritime law."],
        ["What are the legal requirements for registering a vessel in India?"],
    ],
    cache_examples=False,
)

# Launch the Gradio app
if __name__ == "_main_":
    demo.launch()