from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel
import torch
from tqdm import tqdm
import json

# Load model and tokenizer
model_id = "yuhkis/llm-jp-3-13b-finetune"
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,
    device_map="auto",
    token=HF_TOKEN
)
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, token=HF_TOKEN)

# Generate outputs
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=100,
            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"], "output": output})

# Save results to JSONL file
with open("results.jsonl", 'w', encoding='utf-8') as f:
    for result in results:
        json.dump(result, f, ensure_ascii=False)
        f.write('\n')

Output Format Specification

Required fields in the JSONL output:

task_id: Task identifier (integer)
output: Generated response (string)

Example output format:

{"task_id": 0, "output": "応答テキスト"}

Note: While additional fields (e.g., input, eval_aspect) may be included, only task_id and output are required for submission.


### Out-of-Scope Use

This model should not be used for:
- Commercial applications due to license restrictions
- Critical decision-making without human oversight
- Applications requiring strict reliability guarantees

## Bias, Risks, and Limitations

- The model inherits biases from its training data
- Output quality may vary depending on input complexity
- The model should not be used for making critical decisions without human oversight

### Recommendations

Users should be aware of the model's limitations and verify outputs when used in applications.

## Training Details

### Training Data

- Dataset: Ichikara Instruction Dataset

### Training Procedure 

- **Training regime:** bf16 mixed precision
- **Library:** 🤗 Transformers
- **Optimization:** LoRA (Low-Rank Adaptation)

## Technical Specifications

### Model Architecture

- Base model: LLM-jp-3-13b
- Adaptation method: LoRA

## Citation

**BibTeX:**
```bibtex
@misc{shiraishi2024llm,
    title={LLM-jp-3-13b-finetune: Instruction-tuned Japanese Language Model},
    author={Yuhki Shiraishi},
    year={2024},
    publisher={Hugging Face},
    howpublished={\url{https://huggingface.co/yuhkis/llm-jp-3-13b-finetune}}
}

Base Model Citation:

bibtex

@misc{llm-jp2024,
    title={LLM-jp-3: Large Language Model for Japanese},
    author={LLM-jp Project Team},
    year={2024},
    publisher={Hugging Face},
    howpublished={\url{https://huggingface.co/llm-jp/llm-jp-3-13b}}
}

Training Data Citation:

関根聡, 安藤まや, 後藤美知子, 鈴木久美, 河原大輔, 井之上直也, 乾健太郎. 
ichikara-instruction: LLMのための日本語インストラクションデータの構築. 
言語処理学会第30回年次大会(2024)

Model Card Contact

Primary Contact:

Name: Yuhki Shiraishi
GitHub: @yuhkis

For questions regarding this model, please open an issue in the GitHub repository or contact via HuggingFace discussion forum.

Please include "LLM-jp-3-13b-finetune" in the subject line of any correspondence.