import os
import random
import gradio as gr
from langchain_core.prompts import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
    MessagesPlaceholder,
)
from langchain_core.messages import SystemMessage
from langchain.chains.conversation.memory import ConversationBufferWindowMemory
from langchain_groq import ChatGroq
from langchain_google_genai import ChatGoogleGenerativeAI

# Initialize Groq Langchain chat object and conversation
groq_chat = ChatGroq(
    groq_api_key=os.environ.get("GROQ_API_KEY"),
)
# Initialize Google Langchain chat object and conversation
google_chat = ChatGoogleGenerativeAI(
    api_key=os.environ.get("GOOGLE_API_KEY"),
)


# Initialize memory to manages the chat history, 
# ensuring the AI remembers the specified number of history messages, in this case 8.
memory = ConversationBufferWindowMemory(k=8, memory_key="chat_history", return_messages=True)


def generate_response(user_input, history, model, temperature, max_tokens, top_p, seed):
    print( "Model =", model)

    if model.startswith("gemini"):
        chat = google_chat
        chat.model = model
    else:
        chat = groq_chat
        chat.model_name = model

    prompt = ChatPromptTemplate.from_messages(
        [
            # This is the persistent system prompt, sets the initial context for the AI.
            SystemMessage(content='You are a helpful AI assistant.'),
            # This placeholder will take care of chat history.
            MessagesPlaceholder(variable_name="chat_history"),
            # This template is where the user's current input will be injected into the prompt.
            HumanMessagePromptTemplate.from_template("{human_input}"),
        ]
    )    

    # Create a conversation sequence using RunnableSequence
    conversation = prompt | chat

    # Load chat_history
    chat_history = memory.load_memory_variables({})["chat_history"]
    
    # The chatbot's answer is generated by sending the full prompt to the LLM
    response = conversation.invoke({"human_input": user_input, "chat_history": chat_history})

    # Update the memory with the new interaction
    memory.save_context({"input": user_input}, {"output": response.content})
    
    return response.content

# Define additional inputs and examples if needed
additional_inputs = [
    gr.Dropdown(choices=["llama-3.1-70b-versatile", "llama-3.1-8b-instant", "llama3-70b-8192", "llama3-8b-8192", "mixtral-8x7b-32768", "gemma2-9b-it", "gemma-7b-it","gemini-1.5-pro", "gemini-1.5-flash", "gemini-1.5-flash-8b", "gemini-2.0-flash-exp"], value="llama-3.1-70b-versatile", label="Model"),
    gr.Slider(minimum=0.0, maximum=1.0, step=0.01, value=0.5, label="Temperature", info="Controls diversity of the generated text. Lower is more deterministic, higher is more creative."),
    gr.Slider(minimum=1, maximum=8000, step=1, value=8000, label="Max Tokens", info="The maximum number of tokens that the model can process in a single response.<br>Maximums: 8k for gemma 7b it, gemma2 9b it, llama 7b & 70b, 32k for mixtral 8x7b, 132k for llama 3.1."),
    gr.Slider(minimum=0.0, maximum=1.0, step=0.01, value=0.5, label="Top P", info="A method of text generation where a model will only consider the most probable next tokens that make up the probability p."),
    gr.Number(precision=0, value=0, label="Seed", info="A starting point to initiate generation, use 0 for random")
]

example1 = [
            ["What's the distance from Tokyo to New York?"],    
            ["What to San Francisco?"],        
            ["Then what to Beijing?"],
            ["And what to Kyoto?"],
            ["What from Beijing to New York?"]
        ]

# Create the Gradio interface
interface = gr.ChatInterface(
    fn=generate_response, 
    chatbot=gr.Chatbot(show_label=False, show_share_button=False, show_copy_button=True, likeable=True, layout="panel"),
    additional_inputs=additional_inputs,
    examples=example1,
    cache_examples=False,
)

# Launch the app
interface.launch()