import gradio as gr
import torch
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    TextIteratorStreamer,
)
import os
from threading import Thread
import spaces
import time
import subprocess

subprocess.run(
    "pip install flash-attn --no-build-isolation",
    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
    shell=True,
)


model = AutoModelForCausalLM.from_pretrained(
    "microsoft/Phi-3-small-8k-instruct",
    torch_dtype="auto", 
    trust_remote_code=True,
)
tok = AutoTokenizer.from_pretrained("microsoft/Phi-3-small-8k-instruct",trust_remote_code=True,)
terminators = [
    tok.eos_token_id,
]

if torch.cuda.is_available():
    device = torch.device("cuda")
    print(f"Using GPU: {torch.cuda.get_device_name(device)}")
else:
    device = torch.device("cpu")
    print("Using CPU")

model = model.to(device)


@spaces.GPU(duration=60)
def chat(message, history,system_prompt, temperature, do_sample, max_tokens, top_k, repetition_penalty, top_p):
    chat = [
        {"role": "assistant", "content": system_prompt}
    ]
    for item in history:
        chat.append({"role": "user", "content": item[0]})
        if item[1] is not None:
            chat.append({"role": "assistant", "content": item[1]})
    chat.append({"role": "user", "content": message})
    messages = tok.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
    model_inputs = tok([messages], return_tensors="pt").to(device)
    streamer = TextIteratorStreamer(
        tok, timeout=20.0, skip_prompt=True, skip_special_tokens=True
    )
    generate_kwargs = dict(
        model_inputs,
        streamer=streamer,
        max_new_tokens=max_tokens,
        do_sample=True,
        temperature=temperature,
        eos_token_id=terminators,
        top_k=top_k,
        repetition_penalty=repetition_penalty,
        top_p=top_p
    )

    if temperature == 0:
        generate_kwargs["do_sample"] = False

    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    partial_text = ""
    for new_text in streamer:
        partial_text += new_text
        yield partial_text

    yield partial_text


demo = gr.ChatInterface(
    fn=chat,
    examples=[["Write me a poem about Machine Learning."],
              ["write fibonacci sequence in python"],
              ["who won the world cup in 2018?"],
              ["when was the first computer invented?"],
              ],
    additional_inputs_accordion=gr.Accordion(
        label="⚙️ Parameters", open=False, render=False
    ),
    additional_inputs=[
        gr.Textbox("Perform the task to the best of your ability.", label="System prompt"),
        gr.Slider(
            minimum=0, maximum=1, step=0.1, value=0.9, label="Temperature", render=False
        ),
        gr.Checkbox(label="Sampling", value=True),
        gr.Slider(
            minimum=128,
            maximum=4096,
            step=1,
            value=512,
            label="Max new tokens",
            render=False,
        ),
        gr.Slider(1, 80, 40, label="Top K sampling"),
        gr.Slider(0, 2, 1.1, label="Repetition penalty"),
        gr.Slider(0, 1, 0.95, label="Top P sampling"),
    ],
    stop_btn="Stop Generation",
    title="Chat With Phi-3-Small-8k-7b-Instruct",
    description="[microsoft/Phi-3-small-8k-instruct](https://huggingface.co/microsoft/Phi-3-small-8k-instruct)",
    css="footer {visibility: hidden}",
    theme="NoCrypt/miku@1.2.1",
)
demo.launch()