git lfs install
git clone https://huggingface.co/llm-jp/llm-jp-3-13b

import os,sys
os.environ['CUDA_VISIBLE_DEVICES'] = '1'
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    BitsAndBytesConfig,
)
from peft import PeftModel
import torch
from tqdm import tqdm
import json, re
from pathlib import Path

def inference(model_id, adapter_id, datasets_path, token):
    
    # QLoRA config
    bnb_config = BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_quant_type="nf4",
        bnb_4bit_compute_dtype=torch.bfloat16,
    )    

    # Load model
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        quantization_config=bnb_config,
        device_map="auto",
        token = token
    )
    
    # Load tokenizer
    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, token = HF_TOKEN)    
    
    # 元のモデルにLoRAのアダプタを統合。
    model = PeftModel.from_pretrained(model, adapter_id, token = HF_TOKEN)
    model.eval()  
    
    # datasets
    datasets = []
    with open(datasets_path, "r") as f:
        item = ""
        for line in f:
            line = line.strip()
            item += line
            if item.endswith("}"):
                datasets.append(json.loads(item))
                item = ""
    
    # inference
    results = []
    for data in tqdm(datasets):

        input = data["input"]

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

        tokenized_input = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt").to(model.device)
        attention_mask = torch.ones_like(tokenized_input)
        with torch.no_grad():
            outputs = model.generate(
                tokenized_input,
                attention_mask=attention_mask,
                max_new_tokens=4096,
                do_sample=False,
                repetition_penalty=1.2,
                pad_token_id=tokenizer.eos_token_id
            )[0]
        output = tokenizer.decode(outputs[tokenized_input.size(1):], skip_special_tokens=True)

        results.append({"task_id": data["task_id"], "input": input, "output": output})            
    return results

if __name__ == '__main__':
    # hugging face 
    HF_TOKEN = "write your huggingface token"
    
    # model_id
    model_id = Path("modules")/'llm-jp-3-13b'
    
    # adapter_id
    adapter_id = f'ShunShimura/llm-jp-3-13b-finetune'
    
    # datasets
    datasets_path = Path(".")/'elyza-tasks-100-TV_0.jsonl'
    
    # inference
    results = inference(model_id, adapter_id, datasets_path, HF_TOKEN)
    
    jsonl_id = re.sub(".*/", "", adapter_id)
    with open(f"./{jsonl_id}-outputs.jsonl", 'w', encoding='utf-8') as f:
        for result in results:
            json.dump(result, f, ensure_ascii=False)  # ensure_ascii=False for handling non-ASCII characters
            f.write('\n')