πŸš€ Update News

  • 2025-10-13: Official release of KORMo-10B-sft.

πŸ’‘ About KORMo

KORMo-10B is a 10.8B parameter fully open LLM capable of handling both Korean and English.
The model, training code, and training data are all fully open, allowing anyone to reproduce and extend them.

  • Model Size: 10.8B parameters
  • Languages: Korean / English
  • Training Data: Synthetic data + public datasets (approximately 3T tokens)
  • License: Apache 2.0
KORMoλŠ” λΉ„μ˜μ–΄κΆŒ 졜초의 Fully Open Source LLM으둜, 곡읡적 ν™œμš©μ„ λͺ©ν‘œλ‘œ νƒ„μƒν–ˆμŠ΅λ‹ˆλ‹€.
μš°λ¦¬λŠ” λˆ„κ΅¬λ‚˜ 세계 μˆ˜μ€€μ˜ μ–Έμ–΄λͺ¨λΈμ„ 직접 λ§Œλ“€κ³  λ°œμ „μ‹œν‚¬ 수 μžˆλŠ” ν™˜κ²½μ„ λ§Œλ“€κ³ μž ν•©λ‹ˆλ‹€.
KORMo의 μ£Όμš” νŠΉμ§•μ€ λ‹€μŒκ³Ό κ°™μŠ΅λ‹ˆλ‹€:

1. From scratch ν•™μŠ΅μœΌλ‘œ μ„€κ³„λœ 10BκΈ‰ ν•œβ€“μ˜ μΆ”λ‘  μ–Έμ–΄λͺ¨λΈμž…λ‹ˆλ‹€.
2. ν•™μŠ΅ 데이터, μ½”λ“œ, λͺ¨λΈ μ²΄ν¬ν¬μΈνŠΈμ™€ νŠœν† λ¦¬μ–Όμ„ 100% κ³΅κ°œν•˜μ—¬, λˆ„κ΅¬λ‚˜ SOTA에 κ·Όμ ‘ν•œ λͺ¨λΈμ„ 직접 μž¬ν˜„ν•˜κ³  ν™•μž₯ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
3. 총 3.7T 토큰 규λͺ¨μ˜ ν•™μŠ΅ 데이터λ₯Ό κ³΅κ°œν•©λ‹ˆλ‹€. 특히 μ§€κΈˆκΉŒμ§€ ν•œ λ²ˆλ„ 곡개된 적 μ—†λŠ” μ΄ˆκ³ ν’ˆμ§ˆ μ „μ£ΌκΈ° ν•œκ΅­μ–΄ 데이터(μ‚¬μ „ν•™μŠ΅, μ‚¬ν›„ν•™μŠ΅, μΌλ°˜ν˜•, μΆ”λ‘ ν˜•, κ°•ν™”ν•™μŠ΅ λ“±)λ₯Ό μ œκ³΅ν•©λ‹ˆλ‹€.
4. 이 λͺ¨λ“  μž‘μ—…μ€ KAIST λ¬Έν™”κΈ°μˆ λŒ€ν•™μ› MLPμ—°κ΅¬μ‹€μ˜ 학뢀·석사생 8λͺ…이 ν˜‘λ ₯ν•˜μ—¬ μ§„ν–‰ν–ˆμœΌλ©°, 45μž₯에 λ‹¬ν•˜λŠ” λ…Όλ¬ΈμœΌλ‘œ μ •λ¦¬ν–ˆμŠ΅λ‹ˆλ‹€.

μ§€κΈˆκΉŒμ§€ ν•œκ΅­μ–΄ λͺ¨λΈμ„ 써보면, 벀치마크 μ μˆ˜λŠ” 쒋은데 μ‹€μ‚¬μš©μ—μ„œλŠ” μ–΄λ”˜κ°€ μ΄μƒν•˜κ±°λ‚˜,
νŠœλ‹λ§Œ ν•˜λ©΄ λͺ¨λΈμ΄ λ§κ°€μ§€λŠ” κ²½ν—˜μ„ ν•˜μ…¨μ„ κ²λ‹ˆλ‹€. λ‹΅λ‹΅ν•˜μ…¨μ£ ?

KORMoλŠ” 그런 문제λ₯Ό μ •λ©΄μœΌλ‘œ ν•΄κ²°ν•©λ‹ˆλ‹€.
λͺ¨λ“  쀑간 λͺ¨λΈκ³Ό μ‚¬ν›„ν•™μŠ΅ 데이터λ₯Ό ν•¨κ»˜ κ³΅κ°œν•˜κΈ° λ•Œλ¬Έμ—, μ‚¬μš©μžλŠ” 베이슀 λͺ¨λΈ μœ„μ— μžμ‹ λ§Œμ˜ 데이터λ₯Ό μ–Ήμ–΄ μ›ν•˜λŠ” λ°©ν–₯으둜 κ°•ν™”ν•™μŠ΅Β·νŠœλ‹μ„ μ§„ν–‰ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
πŸ‘‰ "쒋은 ν•œκ΅­μ–΄ λͺ¨λΈμ„ κ°–κ³  μ‹Άλ‹€λ©΄, 이제 직접 λ§Œλ“€μ–΄λ³΄μ„Έμš”. μ½”λž© 무료 GPUλ‘œλ„ νŠœλ‹λ©λ‹ˆλ‹€! πŸ€—"

πŸ”— Links


πŸ“ˆ Benchmark Performance

πŸ“Š Quantitative Evaluation

Benchmark KORMo-10B smolLM3-3B olmo2-7B olmo2-13B kanana1.5-8B qwen3-8B llama3.1-8B gemma3-4B gemma3-12B
πŸ‡ΊπŸ‡Έ English Benchmarks
arc_challenge 58.96 55.55 59.13 61.01 56.48 63.82 54.61 53.58 63.82
arc_easy 85.48 83.21 85.06 86.57 82.74 87.50 84.01 82.83 87.37
boolq 83.46 82.17 84.50 86.48 84.53 87.71 81.87 80.70 86.61
copa 93.00 91.00 92.00 93.00 88.00 92.00 93.00 89.00 95.00
gpqa_main 30.13 26.79 26.34 29.24 29.24 30.13 23.44 30.13 35.71
hellaswag 60.25 56.78 61.52 65.02 59.93 59.54 60.96 57.56 63.67
mmlu 67.96 61.37 62.81 66.85 63.73 76.95 65.03 59.60 73.58
mmlu_global 63.44 57.52 59.88 63.99 60.21 75.05 61.30 57.23 70.23
mmlu_pro 40.18 34.94 27.29 32.50 34.93 56.58 36.23 27.79 37.07
mmlu_redux 69.00 62.95 63.53 68.37 65.88 78.19 65.86 60.86 75.25
openbookqa 39.00 36.40 39.00 39.60 36.80 39.20 39.00 37.00 40.20
piqa 81.12 78.45 80.79 82.64 80.30 79.05 80.90 79.49 82.59
social_iqa 52.81 50.72 55.89 57.57 57.01 56.96 53.12 51.84 56.45
English Avg. 63.45 59.83 61.36 64.06 61.52 67.90 61.49 59.05 66.73
πŸ‡°πŸ‡· Korean Benchmarks
click 55.29 46.97 37.79 41.80 62.76 60.70 49.22 49.62 62.21
csatqa 38.00 26.67 19.33 24.67 44.67 52.00 28.67 28.67 31.33
haerae 68.29 55.82 31.62 37.58 80.75 67.19 53.25 60.68 74.34
k2_eval 84.89 75.23 49.54 63.43 84.72 84.72 76.62 76.39 85.42
kobest 75.05 69.13 57.27 59.02 81.93 80.05 70.55 69.33 77.70
kobalt 22.86 15.86 11.43 13.14 26.29 26.57 17.43 15.57 23.86
kmmlu 46.48 38.52 33.05 31.24 48.86 56.93 40.75 39.84 51.60
mmlu_global (ko) 55.16 44.15 34.00 36.95 52.65 61.95 46.34 46.33 59.68
kr_clinical_qa 77.32 53.97 48.33 46.22 65.84 80.00 63.54 60.00 77.22
Korean Avg. 58.15 47.37 35.82 39.34 60.94 63.35 49.60 49.60 60.37

πŸ“ Qualitative Evaluation (LLM-as-a-Judge)

Benchmark KORMo-10B smolLM3-3B olmo2-7B olmo2-13B kanana1.5-8B qwen3-8B llama3.1-8B exaone3.5-8B gemma3-12B
MT-Bench (EN) 8.32 7.15 7.32 7.64 8.45 8.70 6.32 8.15 8.70
KO-MT-Bench (KO) 8.54 - - - 8.02 8.16 4.27 8.13 8.51
LogicKor (KO) 8.96 - - - 8.94 8.63 6.45 9.20 8.46
Average 8.61 - - - 8.47 8.50 5.68 8.49 8.56

πŸ“¦ Installation

git clone https://github.com/MLP-Lab/KORMo-tutorial.git
cd KORMo-tutorial
bash setup/create_uv_venv.sh
source .venv_kormo/bin/activate

πŸš€ Inference Example

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "KORMo-Team/KORMo-10B-sft"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)

messages = [
    {"role": "user", "content": "What happens inside a black hole?"}
]

chat_prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=False
)

inputs = tokenizer(chat_prompt, return_tensors="pt").to(model.device)

with torch.inference_mode():
    output_ids = model.generate(
        **inputs,
        max_new_tokens=1024,
    )

response = tokenizer.decode(output_ids[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
print("Assistant:", response)

🧠 Enabling Thinking Mode

If you want to enable the thinking mode, simply set enable_thinking=True:

chat_prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True
)

Limitation

The model has not yet been safety-tuned or preference-aligned, which may lead to suboptimal performance or undesired repetitions in complex reasoning tasks.

Contact

Acknowledgments

  • This work was supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government(MSIT) (RS-2025-02653113, High-Performance Research AI Computing Infrastructure Support at the 2 PFLOPS Scale)

Citation

@misc{KORMo,
  author = {Minjun Kim, Hyeonseok Lim, Hangyeol Yoo, Inho Won, Seungwoo Song, Minkyung Cho, Junghun Yuk, Changsu Choi, Dongjae Shin, Huije Lee, Hoyun Song, Alice Oh, and KyungTae Lim},
  title = {KORMo: Korean Open Reasoning Model for Everyone},
  year = {2025},
  publisher = {GitHub},
  journal = {Technical Report},
  paperLink = {\url{https://arxiv.org/abs/2510.09426}},
 },
}
Downloads last month
2,108
Safetensors
Model size
11B params
Tensor type
BF16
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for KORMo-Team/KORMo-10B-sft

Finetunes
3 models
Quantizations
3 models

Space using KORMo-Team/KORMo-10B-sft 1

Collection including KORMo-Team/KORMo-10B-sft