README.md

---
language: ko
pipeline_tag: text-generation
license: llama3.1
---

### 1. Model Description
- KONI (KISTI Open Natural Intelligence) is a specialized large language model (LLM) developed by the Korea Institute of Science and Technology Information (KISTI). This model is specifically designed for science and technology, making it highly effective for tasks in these fields.

### 2. Key Features
- **Specialized in Science and Technology:** The model is explicitly trained on a vast and specialized corpus of scientific and technological data.
- **Enhanced Reasoning Performance:** This version of KONI demonstrates significantly enhanced reasoning performance compared to its instruction-tuned version released in October 2024.
- **Base Model:** The base model for KONI-Llama3.1-8B-R-Preview-20250320 is KONI-Llama3.1-8B-Merged-20240830, which is a merger of Meta-Llama-3-8B and KONI-Llama3.1-8B-20240824.
- **Alignment:** SFT (Supervised Fine-Tuning) is applied.

### 3. Data
- Approximately 5k SFT data for reasoning.
- **SFT Data:** The SFT data includes both internally generated data and publicly available CoT data on Hugging Face, translated into Korean where necessary.

### 4. Benchmark Results
The model performance will be released later.

### 5. How to use the model
```python
import transformers
import torch

model_id = "KISTI-KONI/KONI-Llama3.1-8B-R-Preview-20250320"

pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

pipeline.model.eval()

instruction = "안녕? 너는 누구야?"

messages = [
   {"role": "user", "content": f"{instruction}"}
    ]

prompt = pipeline.tokenizer.apply_chat_template(
        messages, 
        tokenize=False, 
        add_generation_prompt=True
)

terminators = [
    pipeline.tokenizer.eos_token_id,
    pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = pipeline(
    prompt,
    max_new_tokens=2048,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.7,
    top_p=0.9
)

print(outputs[0]["generated_text"][len(prompt):])
```
```
<think>
Alright, let's tackle this question. The user is asking, "Hello? Who are you?" in a friendly and informal manner. They probably want a response that's equally casual and welcoming.

First, I should make sure to address them directly with a warm greeting. Maybe start with "Hey there!" or "Hi!" to set a friendly tone. Then, since they're asking who I am, I need to introduce myself. Keep it brief and light. Maybe mention that I'm an AI assistant trained based on LlaMA3.1 and created by KISTI to be helpful and honest.

I should also note that I'm here to engage in substantive conversations and provide analysis on complex subjects. That shows I can handle deeper topics if needed. It's important to be honest and not pretend to be something I'm not, so mentioning that I don't know things if I don't know them is good.

Finally, end with a friendly sign-off like "Take care!" or "Talk to you later!" to keep the tone upbeat. Make sure the whole response is easy to read and flows naturally. Avoid jargon so it's accessible. Let me put this all together in a way that feels approachable and honest.
</think>
<solution>
안녕하세요! 저는 KISTI에서 개발한 대화형 AI 어시스턴트 LlaMA3.1 기반의 KONI입니다. 저는 복잡한 주제에 대해 심도 있는 대화를 나누고 분석을 제공하도록 훈련받았습니다. 저는 도움을 주기 위해 여기 있습니다! 알려주세요~ 저는 무엇을 도와드릴 수 있을까요? 제가 모르는 것도 있으니, 제가 잘 모르는 것은 정직하게 알려드리겠습니다. 😊

---
*더 깊이 있는 대화나 특정 주제에 대해 논의하고 싶으시면 언제든지 말씀해 주세요!*

---
*추신: 질문을 더 잘 이해하도록 도와주시고 싶으시면, 더 많은 정보가 필요하시면 말씀해 주세요! 😊*
```

### 6. Citation
**Language Model**
```text
@article{KISTI-KONI/KONI-Llama3.1-8B-R-Preview-20250320,
  title={KISTI-KONI/KONI-Llama3.1-8B-R-Preview-20250320},
  author={KISTI},
  year={2025},
  url={https://huggingface.co/KISTI-KONI/KONI-Llama3.1-8B-R-Preview-20250320}
}
```
  
### 7. Contributors
- KISTI, Large-scale AI Research Center

### 8. Special Thanks
- [@GUIJIN SON](https://huggingface.co/amphora)
- [@beomi](https://huggingface.co/beomi)
- [@kuotient](https://huggingface.co/kuotient)
- KyungTae Lim

### 8. Acknowledgement
- This research was supported by Korea Institute of Science and Technology Information(KISTI).
- This work was supported by the National Supercomputing Center with supercomputing resources including technical support (KISTI).

### 9. References
- https://huggingface.co/meta-llama/Meta-Llama-3.1-8B
- https://huggingface.co/meta-llama/meta-llama/Meta-Llama-3.1-8B-Instruct