Instructions to use cooler8/yejin-korean-3b-v2-base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cooler8/yejin-korean-3b-v2-base with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="cooler8/yejin-korean-3b-v2-base")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cooler8/yejin-korean-3b-v2-base")
model = AutoModelForCausalLM.from_pretrained("cooler8/yejin-korean-3b-v2-base", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use cooler8/yejin-korean-3b-v2-base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cooler8/yejin-korean-3b-v2-base"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cooler8/yejin-korean-3b-v2-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cooler8/yejin-korean-3b-v2-base

SGLang

How to use cooler8/yejin-korean-3b-v2-base with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "cooler8/yejin-korean-3b-v2-base" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cooler8/yejin-korean-3b-v2-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "cooler8/yejin-korean-3b-v2-base" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cooler8/yejin-korean-3b-v2-base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cooler8/yejin-korean-3b-v2-base with Docker Model Runner:
```
docker model run hf.co/cooler8/yejin-korean-3b-v2-base
```

yejin-korean-3b-v2-base

100% from-scratch 사전학습된 한국어 3B 파운데이션 모델. 외부 공개 모델 가중치를 전혀 사용하지 않았습니다.

파라미터: 3.02B (hidden 3072, 28 layers, 24 heads, 8 KV heads, GQA 3:1)
컨텍스트 4096, RoPE theta 500000, QK-Norm, tied embeddings
토크나이저: 독자 64K Byte-level BPE (한국어 특화)
학습: 8x NVIDIA H200, 34000 steps, 약 73B 토큰 (155GB 한국어 코퍼스: AI Hub, CulturaX, 위키백과, 교과서 등)
학습 손실: ~2.0

HF 아키텍처 클래스에 관하여

학습 모델은 QK-Norm(헤드별 RMSNorm, RoPE 이전) 을 사용합니다. HF LlamaForCausalLM 에는 QK-Norm이 없어 동일 구조를 지원하는 Qwen3ForCausalLM 클래스로 로드합니다. 가중치는 Qwen 과 무관한 독자 학습 결과이며, 클래스는 모델 구조(연산 그래프) 호환을 위해서만 사용됩니다. vLLM / llama.cpp 에서 그대로 로드 가능합니다.

사용

from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("cooler8/yejin-korean-3b-v2-base")
model = AutoModelForCausalLM.from_pretrained("cooler8/yejin-korean-3b-v2-base", torch_dtype="bfloat16", device_map="auto")
ids = tok("대한민국의 수도는", return_tensors="pt").to(model.device)
print(tok.decode(model.generate(**ids, max_new_tokens=50)[0]))

이 모델은 base 모델입니다 (지시문 튜닝 없음). SFT/DPO 버전: cooler8/yejin-korean-3b-v2-sft, cooler8/yejin-korean-3b-v2-dpo.

변환: step 34000 체크포인트, 2026-09-12 02:27 UTC

Downloads last month: 358

Safetensors

Model size

3B params

Tensor type

BF16