import asyncio
import json
import os

import aiohttp
import gradio as gr
import numpy as np
import spaces
from huggingface_hub import InferenceClient

import random
import torch
from huggingface_hub import AsyncInferenceClient
from transformers import LlamaTokenizer, LlamaForCausalLM, AutoTokenizer


async def query_llm(payload, model_name):
    headers = {"Authorization": f"Bearer {os.getenv('HF_TOKEN')}"}
    async with aiohttp.ClientSession() as session:
        async with session.post(f"https://api-inference.huggingface.co/models/{model_name}", headers=headers,
                                json=payload) as response:
            return await response.json()


async def generate_mistral_7bvo1(system_input, user_input):
    client = AsyncInferenceClient(
        "mistralai/Mistral-7B-Instruct-v0.1",
        token=os.getenv('HF_TOKEN'),
    )

    async for message in await client.chat_completion(
            messages=[
                {"role": "system", "content": system_input},
                {"role": "user", "content": user_input}, ],
            max_tokens=256,
            stream=True,
    ):
        yield message.choices[0].delta.content


async def generate_gpt2(system_input, user_input):
    output = await query_llm({
        "inputs": (inputs:=f"{system_input}\n{user_input}"),
    }, "openai-community/gpt2")
    yield output[0]["generated_text"].replace(inputs, '')


async def generate_llama2(system_input, user_input):
    client = AsyncInferenceClient(
        "meta-llama/Llama-2-7b-chat-hf",
        token=os.getenv('HF_TOKEN')
    )
    async for message in await client.chat_completion(
            messages=[
                {"role": "system", "content": system_input},
                {"role": "user", "content": user_input}, ],
            max_tokens=256,
            stream=True,
    ):
        yield message.choices[0].delta.content


@spaces.GPU(duration=120)
async def generate_openllama(system_input, user_input):
    model_path = 'openlm-research/open_llama_3b_v2'
    tokenizer = LlamaTokenizer.from_pretrained(model_path)
    model = LlamaForCausalLM.from_pretrained(
        model_path, torch_dtype=torch.float16, device_map='cuda',
    )
    # model = model.to("cuda")
    input_text = f"{system_input}\n{user_input}"
    input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to("cuda")
    output = model.generate(input_ids, max_length=128)
    return tokenizer.decode(output[0], skip_special_tokens=True)


@spaces.GPU(duration=120)
async def generate_bloom(system_input, user_input):
    model_path = 'bigscience/bloom-7b1'
    tokenizer = AutoTokenizer.from_pretrained(model_path)
    model = LlamaForCausalLM.from_pretrained(
        model_path, torch_dtype=torch.float16, device_map='cuda',
    )
    input_text = f"{system_input}\n{user_input}"
    input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to("cuda")
    output = model.generate(input_ids, max_length=128)
    return tokenizer.decode(output[0], skip_special_tokens=True)



async def generate_llama3(system_input, user_input):
    client = AsyncInferenceClient(
        "meta-llama/Meta-Llama-3.1-8B-Instruct",
        token=os.getenv('HF_TOKEN')
    )
    try:
        async for message in await client.chat_completion(
                messages=[
                    {"role": "system", "content": system_input},
                    {"role": "user", "content": user_input}, ],
                max_tokens=256,
                stream=True,
        ):
            yield message.choices[0].delta.content
    except json.JSONDecodeError:
        pass