import os
from threading import Thread
from typing import Iterator

import gradio as gr
import spaces
import torch
import json
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

DESCRIPTION = """\
Shakti is a 2.5 billion parameter language model specifically optimized for resource-constrained environments such as edge devices, including smartphones, wearables, and IoT systems. With support for vernacular languages and domain-specific tasks, Shakti excels in industries such as healthcare, finance, and customer service.
For more details, please check [here](https://arxiv.org/pdf/2410.11331v1).
"""

MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "2048"))

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

# Model configurations
model_options = {
    "Shakti-100M": "SandLogicTechnologies/Shakti-100M",
    "Shakti-250M": "SandLogicTechnologies/Shakti-250M",
    "Shakti-2.5B": "SandLogicTechnologies/Shakti-2.5B"
}

# Initialize tokenizer and model variables
tokenizer = None
model = None

def load_model(selected_model: str):
    global tokenizer, model
    model_id = model_options[selected_model]
    tokenizer = AutoTokenizer.from_pretrained(model_id, token=os.getenv("SHAKTI"))
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        device_map="auto",
        torch_dtype=torch.bfloat16,
        token=os.getenv("SHAKTI")
    )
    model.eval()

# Initial model load (default to 2.5B)
load_model("Shakti-2.5B")

@spaces.GPU(duration=90)
def generate(
    message: str,
    chat_history: list[tuple[str, str]],
    max_new_tokens: int = 1024,
    temperature: float = 0.6,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
) -> Iterator[str]:
    conversation = []
    for user, assistant in chat_history:
        conversation.extend(
            [
                json.loads(os.getenv("PROMPT")),
                {"role": "user", "content": user},
                {"role": "assistant", "content": assistant},
            ]
        )
    conversation.append({"role": "user", "content": message})

    input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt")
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    input_ids = input_ids.to(model.device)

    streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs)

def update_examples(selected_model):
    if selected_model == "Shakti-100M":
        return [["Tell me a story"],
            ["Write a short poem on Rose"],
            ["What are computers"]]
    elif selected_model == "Shakti-250M":
        return [["Can you explain the pathophysiology of hypertension and its impact on the cardiovascular system?"],
            ["What are the potential side effects of beta-blockers in the treatment of arrhythmias?"],
            ["What foods are good for boosting the immune system?"],
			["What is the difference between a stock and a bond?"],
			["How can I start saving for retirement?"],
			["What are some low-risk investment options?"],
			["What is a power of attorney and when is it used?"],
			["What are the key differences between a will and a trust?"],
			["How do I legally protect my business name?"]]
    else:
        return [["Tell me a story"], ["write a short poem which is hard to sing"], ['मुझे भारतीय इतिहास के बारे में बताएं']]

def on_model_select(selected_model):
    load_model(selected_model)  # Load the selected model
    return update_examples(selected_model)  # Return new examples based on the selected model


chat_interface = gr.ChatInterface(
    fn=generate,
    additional_inputs=[
        gr.Slider(
            label="Max new tokens",
            minimum=1,
            maximum=MAX_MAX_NEW_TOKENS,
            step=1,
            value=DEFAULT_MAX_NEW_TOKENS,
        ),
        gr.Slider(
            label="Temperature",
            minimum=0.1,
            maximum=4.0,
            step=0.1,
            value=0.6,
        ),
    ],
    stop_btn=None,
    examples=update_examples("Shakti-2.5B"),  # Set initial examples for 2.5B model
    cache_examples=False,
)

with gr.Blocks(css="style.css", fill_height=True) as demo:
    gr.Markdown(DESCRIPTION)
    gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")

    # Dropdown for model selection
    model_dropdown = gr.Dropdown(
        label="Select Model",
        choices=["Shakti-100M", "Shakti-250M", "Shakti-2.5B"],
        value="Shakti-2.5B",
        interactive=True,
    )

    # Function to handle model change and update examples dynamically
    model_dropdown.change(on_model_select, inputs=model_dropdown, outputs=[chat_interface])

    chat_interface.render()

if __name__ == "__main__":
    demo.queue(max_size=20).launch()