# 必要なライブラリをインストール
!pip uninstall unsloth -y
!pip install --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
!pip install --upgrade torch
!pip install --upgrade xformers


# 必要なライブラリを読み込み
from unsloth import FastLanguageModel
import json
from tqdm import tqdm
import re


# モデルをロード
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "mnm373/gemma-2-9b-it-v3_lora",
    load_in_4bit = True,
    trust_remote_code=True,
)
FastLanguageModel.for_inference(model) # Enable native 2x faster inference

# データセットの読み込み
# 事前にデータをアップロードしてください
datasets = []
with open("./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 = ""

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

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

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

    outputs = model.generate(**inputs, max_new_tokens=1024, use_cache=True, do_sample=False, repetition_penalty=1.2)
    prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答\n')[-1]

    # 不要なフレーズを削除
    if prediction.startswith("こんにちは！"):
        prediction = prediction.lstrip("こんにちは！")
    if prediction.startswith("もちろんです！"):
        prediction = prediction.lstrip("もちろんです！")

    phrases_to_remove = [
        "ユーモアを交えてお答えしますね。",
        "ユーモアを交えつつお答えしますね。"
    ]
    for phrase in phrases_to_remove:
        prediction = prediction.replace(phrase, "")

    # 不要な空白や改行をトリミング
    prediction = prediction.strip()

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


# 結果をjsonlで保存。
json_file_id = re.sub(".*/", "", adapter_id)
with open(f"/content/{json_file_id}_output.jsonl", 'w', encoding='utf-8') as f:
    for result in results:
        json.dump(result, f, ensure_ascii=False)
        f.write('\n')