import json
from tqdm import tqdm
import sys
import os

import openai
from tenacity import (
    retry,
    stop_after_attempt,
    wait_random_exponential,
)  # for exponential backoff
from tqdm import tqdm
from datasets import load_dataset
import torch
from unsloth import FastLanguageModel

def main():
    model_name = "aki916/llm-jp-3-13b-it-v1.2"

    max_seq_length = 2048
    dtype = None
    load_in_4bit = True

    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name = model_name,
        max_seq_length = max_seq_length,
        dtype = dtype,
        load_in_4bit = load_in_4bit,
    )
    FastLanguageModel.for_inference(model)


    # データセットの読み込み。
    datasets = []
    with open("data/elyza-tasks-100-TV_0.jsonl", "r") as f:
        item = ""
        for line in f:
            line = line.strip()
            item += line
            if item.endswith("}"):
                datasets.append(json.loads(item))
                item = ""


    # 推論
    results = []
    for dt in tqdm(datasets):
        input = dt["input"]

        prompt = f"""### 指示\n{input}\n### 回答\n"""

        inputs = tokenizer([prompt], return_tensors = "pt").to(model.device)

        outputs = model.generate(**inputs, max_new_tokens = 512, use_cache = True, do_sample=False, repetition_penalty=1.2)
        prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答')[-1]

        results.append({"task_id": dt["task_id"], "input": input, "output": prediction})

    # jsonlで保存
    with open(f"{model_name.replace('/', '_', -1)}_output.jsonl", 'w', encoding='utf-8') as f:
        for result in results:
            json.dump(result, f, ensure_ascii=False)
            f.write('\n')
    print('finish dump')
    print(f"{model_name.replace('/', '_', -1)}_output.jsonl")


if __name__ == "__main__":
    main()