Instructions to use jwg0830/AX-3.1-Light-sft_v3_1b_0.5 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jwg0830/AX-3.1-Light-sft_v3_1b_0.5 with Transformers:

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

pipe = pipeline("text-generation", model="jwg0830/AX-3.1-Light-sft_v3_1b_0.5")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("jwg0830/AX-3.1-Light-sft_v3_1b_0.5")
model = AutoModelForCausalLM.from_pretrained("jwg0830/AX-3.1-Light-sft_v3_1b_0.5", 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 jwg0830/AX-3.1-Light-sft_v3_1b_0.5 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jwg0830/AX-3.1-Light-sft_v3_1b_0.5"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jwg0830/AX-3.1-Light-sft_v3_1b_0.5",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/jwg0830/AX-3.1-Light-sft_v3_1b_0.5

SGLang

How to use jwg0830/AX-3.1-Light-sft_v3_1b_0.5 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 "jwg0830/AX-3.1-Light-sft_v3_1b_0.5" \
    --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": "jwg0830/AX-3.1-Light-sft_v3_1b_0.5",
		"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 "jwg0830/AX-3.1-Light-sft_v3_1b_0.5" \
        --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": "jwg0830/AX-3.1-Light-sft_v3_1b_0.5",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use jwg0830/AX-3.1-Light-sft_v3_1b_0.5 with Docker Model Runner:
```
docker model run hf.co/jwg0830/AX-3.1-Light-sft_v3_1b_0.5
```

AX-3.1-Light-sft_v3_1b_0.5 (동전, K-AI 등록 버전 3.15)

skt/A.X-3.1-Light 기반 두 파인튜닝 모델 sft_v0_21과 sft_v3_1b_safe를 가중치 평균(model soup, 각 0.5)해서 만든 모델입니다. 데이터 재학습이 아니라 두 모델의 전체 가중치를 원소별로 평균 낸 것입니다.

베이스 모델: skt/A.X-3.1-Light
결합 방법: weight averaging, 0.5×v0.21 + 0.5×v3.1b_safe
입력 모델: sft_v0_21(AI Hub 71857/71874/71610/569/71949, 5,793건, 정답 우선 스키마, 실제 K-AI 평균 0.4326) / sft_v3_1b_safe(위 데이터 + 71875·577 단답형 추가 20,513건, 실제 K-AI 평균 0.4226)
목적: v0.21(CLIcK/Com2 강점)과 v3.1b-safe(KMMLU-Pro/HLE/MuSR 강점)를 가중치 평균으로 결합할 수 있는지 검증
구조 변경: 없음

로컬 평가 결과 (팀원 스위트, B1, 참고용 — K-AI 공식 점수 아님)

벤치마크	base	v0.21	soup-0.5
KMMLU-Pro	37.85%	39.16%	40.11%
CLIcK	63.56%	64.31%	65.71%
HLE	4.40%	4.40%	4.87%
Com2-main	51.00%	51.64%	51.76%
SNU Ko-MuSR	48.40%	56.13%	54.53%
5축 평균	41.04%	43.13%	43.40%

Com2-main·MuSR(Ko)은 이 프로젝트에서 로컬-실제 리더보드 방향이 어긋난 전례가 있어 참고용으로만 사용.

사용 데이터셋 (v0.21 ∪ v3.1b-safe, AI Hub)

ID	데이터셋명	URL
71857	국어 교과 지문형 문제 데이터	https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=71857
71874	전문 의학지식 데이터	https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=71874
71610	금융, 법률 문서 기계독해 데이터	https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=71610
569	행정 문서 대상 기계독해 데이터	https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=569
71949	인과관계 기반 추론 데이터(업사이클링)	https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=71949
71875	필수의료 의학지식 데이터 (단답형만)	https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=71875
577	뉴스 기사 기계독해 데이터 (단답형만)	https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=577

Downloads last month: 49

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for jwg0830/AX-3.1-Light-sft_v3_1b_0.5

Base model

skt/A.X-3.1-Light

Finetuned

(24)

this model