Multiple Choice
Transformers
Safetensors
Korean
qwen3
text-generation
korean
classification
natural-language-inference
function-calling
tool-use
calibrated
Eval Results (legacy)
text-generation-inference
Instructions to use coderecode95/KJH-Judge-4B-ko with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use coderecode95/KJH-Judge-4B-ko with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("coderecode95/KJH-Judge-4B-ko") model = AutoModelForCausalLM.from_pretrained("coderecode95/KJH-Judge-4B-ko", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Download eval/heldout_results.json from coderecode95/KJH-Judge-4B-ko: direct link, hf CLI and curl.
- Browser
- Download file 1.96 kB
-
https://huggingface.co/coderecode95/KJH-Judge-4B-ko/resolve/main/eval/heldout_results.json
- Command line
-
hf download hf://coderecode95/KJH-Judge-4B-ko/eval/heldout_results.json
-
curl -L -o heldout_results.json https://huggingface.co/coderecode95/KJH-Judge-4B-ko/resolve/main/eval/heldout_results.json
1.96 kB
| { | |
| "made": "2026-11-09 13:51", | |
| "protocol": "0-shot, 보기 기호 점수(온도 보정 softmax), 표본 고정 시드", | |
| "results": { | |
| "nsmc": { | |
| "desc": "영화 리뷰 감성 (긍정/부정)", | |
| "source": "e9t/nsmc · ratings_test", | |
| "license": "CC0 1.0", | |
| "n": 1000, | |
| "acc": 0.857, | |
| "ci95": [ | |
| 0.8339, | |
| 0.8773 | |
| ], | |
| "chance": 0.5, | |
| "mean_conf": 0.9491, | |
| "seconds": 120 | |
| }, | |
| "kornli": { | |
| "desc": "문장 관계 3분류 (XNLI 한국어 번역)", | |
| "source": "kakaobrain/kor_nli · xnli/test", | |
| "license": "CC BY-SA 4.0", | |
| "n": 1000, | |
| "acc": 0.745, | |
| "ci95": [ | |
| 0.7171, | |
| 0.771 | |
| ], | |
| "chance": 0.3333, | |
| "mean_conf": 0.8362, | |
| "seconds": 148 | |
| }, | |
| "klue_sts": { | |
| "desc": "두 문장이 같은 뜻인지 (예/아니오)", | |
| "source": "klue/klue · sts/validation (binary-label)", | |
| "license": "CC BY-SA 4.0", | |
| "n": 519, | |
| "acc": 0.8593, | |
| "ci95": [ | |
| 0.8268, | |
| 0.8866 | |
| ], | |
| "chance": 0.5, | |
| "mean_conf": 0.8744, | |
| "seconds": 58 | |
| }, | |
| "unsmile": { | |
| "desc": "악플·혐오 표현 여부", | |
| "source": "smilegate-ai/kor_unsmile · valid (clean 여부)", | |
| "license": "CC BY-NC-ND 4.0", | |
| "n": 1000, | |
| "acc": 0.821, | |
| "ci95": [ | |
| 0.796, | |
| 0.8435 | |
| ], | |
| "chance": 0.5, | |
| "mean_conf": 0.9137, | |
| "seconds": 126 | |
| }, | |
| "kmmlu": { | |
| "desc": "한국어 전문 지식 4지선다 (45과목)", | |
| "source": "HAERAE-HUB/KMMLU · test", | |
| "license": "CC BY-ND 4.0", | |
| "n": 990, | |
| "acc": 0.4263, | |
| "ci95": [ | |
| 0.3958, | |
| 0.4573 | |
| ], | |
| "chance": 0.25, | |
| "mean_conf": 0.5285, | |
| "seconds": 200 | |
| }, | |
| "haerae": { | |
| "desc": "한국 문화·어휘 지식 5지선다", | |
| "source": "HAERAE-HUB/HAE_RAE_BENCH_1.0 · test", | |
| "license": "CC BY-NC-ND 4.0", | |
| "n": 983, | |
| "acc": 0.5615, | |
| "ci95": [ | |
| 0.5303, | |
| 0.5923 | |
| ], | |
| "chance": 0.2084, | |
| "mean_conf": 0.5714, | |
| "seconds": 133 | |
| } | |
| } | |
| } |