mykor/Konan-LLM-OND-gguf

Konan-LLM-OND

Overview

Konan-LLM-OND, a large language model from Konan Technology Inc., is based on Qwen3-4B-Base. It has been specifically optimized for the Korean language through vocabulary expansion, continual pre-training, and instruction tuning to enhance performance and efficiency.

• Languages: Primarily Korean, with support for English.
• Key Features:
  • Expanded Korean Vocabulary: The model's vocabulary has been expanded with additional Korean tokens to improve tokenization efficiency. As a result, Konan-LLM-OND is approximately 30% more token-efficient with Korean input than Qwen3, leading to greater cost-effectiveness and processing speed.
  • Continual Pre-training: The model underwent continual pre-training on a large-scale Korean corpus using an expanded vocabulary. This process enhanced its fundamental understanding and text generation capabilities in Korean.
  • Supervised Fine-Tuning (SFT): The model was fine-tuned on a high-quality Korean instruction dataset to improve its ability to understand and execute a wide variety of real-world tasks.

Benchmark Results

Model Performance (＜ 5B)

Model	Model size	Korean			English
		KMMLU	HRM8K	Ko-IFEval	MMLU	GSM8K	IFEval
Konan-LLM-OND	4.0B	50.6	46.4	68.4	68.8	86.8	73.3
EXAONE-3.5-2.4B-Instruct	2.4B	44.2	31.8	60.5	59.1	81.5	77.7
kanana-1.5-2.1b-instruct-2505	2.1B	32.7	27.2	56.0	52.9	68.8	64.6
Midm-2.0-Mini-Instruct	2.3B	42.4	36.2	66.8	57.4	74.8	68.3
Qwen3-4B(w/o reasoning)	4.0B	-(*)	37.5	68.4	-(*)	83.9	80.0
gemma-3-4b-it	4.3B	38.7	32.7	69.2	59.1	82.2	78.3

Model Performance (≥ 7B)

Model	Model size	Korean			English
		KMMLU	HRM8K	Ko-IFEval	MMLU	GSM8K	IFEval
Konan-LLM-OND	4.0B	50.6	46.4	68.4	68.8	86.8	73.3
A.X-4.0-Light	7.2B	55.3	44.6	71.5	70.6	87.3	81.3
EXAONE-3.5-7.8B-Instruct	7.8B	48.0	39.3	66.8	66.8	91.4	79.9
kanana-1.5-8b-instruct-2505	8.0B	40.4	35.5	71.1	63.1	79.3	76.8
Midm-2.0-Base-Instruct	11.5B	54.2	46.0	75.0	70.2	88.9	79.7
Qwen3-8B(w/o reasoning)	8.1B	-(*)	40.0	70.9	-(*)	84.0	82.8

Note:

• The highest scores are shown in bold.
• (*) Qwen3 models often failed to strictly follow the required answer format in the few-shot setting, which made the scores unreliable. After correcting the evaluation pipeline, we will update the scores.

Benchmark Setup

All benchmarks were executed using the following standardized environment.

• Evaluation Framework: lm-evaluation-harness v0.4.9
• Runtime & Hardware: All models were served with vLLM v0.9.1 on a single NVIDIA GPU.
• Inference Mode: For every benchmark, we invoked the chat_completions API, and scores were computed solely from the generated responses.

Metric Adjustments

• KMMLU was evaluated using the "kmmlu_direct" task in the lm-evaluation-harness.
• MMLU was run with the same configuration as "kmmlu_direct".
• Ko-IFEval was evaluated using the original IFEval protocol, with the dataset sourced from allganize/IFEval-Ko.

Evaluation Protocol

Benchmark	Scoring Method	Few-shot
KMMLU	exact_match	5-shot
HRM8K	mean of hrm8k_gsm8k, hrm8k_ksm, hrm8k_math, hrm8k_mmmlu, hrm8k_omni_math	5-shot
Ko-IFEval	mean of prompt_level_strict_acc, inst_level_strict_acc, prompt_level_loose_acc, inst_level_loose_acc	0-shot
MMLU	exact_match	5-shot
GSM8K	exact_match & flexible-extract	5-shot
IFEval	mean of prompt_level_strict_acc, inst_level_strict_acc, prompt_level_loose_acc, inst_level_loose_acc	0-shot

Quickstart

Konan-LLM-OND is supported in transformers v4.52.0 and later.

pip install transformers>=4.52.0

The code example below shows you how to get the model to generate content based on given inputs.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "konantech/Konan-LLM-OND"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
model.eval()
tokenizer = AutoTokenizer.from_pretrained(model_name)

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "대한민국 수도는?"}
]


input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(
        input_ids,
        max_new_tokens=64,
        do_sample=False,
    )

len_input_prompt = len(input_ids[0])
response = tokenizer.decode(output[0][len_input_prompt:], skip_special_tokens=True)
print(response)
# 대한민국 수도는 서울입니다.

Citation

@misc{Konan-LLM-OND-2025,
  author = {Konan Technology Inc.},
  title = {Konan-LLM-OND},
  year = {2025},
  url = {https://huggingface.co/konantech/Konan-LLM-OND}
}