HyperCLOVA X SEED Think-14B — AI Hub 71949 SFT 3K
This repository contains a standalone BF16 model derived from
naver-hyperclovax/HyperCLOVAX-SEED-Think-14B.
One LoRA adapter was trained on 3,000 examples from AI Hub dataset 71949 and
merged into the pristine base weights.
Model details
- Base model:
naver-hyperclovax/HyperCLOVAX-SEED-Think-14B - Base revision:
9b74e35d4c7e4ffec489f4171273caca8948a2b9 - Architecture:
HyperCLOVAXForCausalLM - Weight format: BF16
safetensors, standalone merged full model - Chat template: official HyperCLOVA X template, preserved from the base
- LoRA: rank
16, alpha32, dropout0.05, biasnone - Target modules:
q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj - Objective: assistant-token-only causal-language-model cross entropy
- Learning rate:
5e-5 - Scheduler: cosine, warmup ratio
0.03, weight decay0 - Training: 3,000 examples, 1 epoch, effective batch size
16 - Per-device batch:
4; gradient accumulation:4 - Maximum sequence length:
4096; precision: BF16; packing:false - Seed:
42; data seed:42 - Public benchmark data: not used
Training data
The training data is AI Hub 71949 인과관계 기반 추론 데이터. The source
material is image-grounded causal-reasoning MCQs; the prepared training rows
convert the underlying causal relation into a text-only 4-choice question with
a text answer, so no image input is required at train or inference time. The
exact prepared input contains 3,000 unique rows spread evenly across 10
causal-relation categories (성장, 가공, 절단, 오염, 작동, 추출, 파손, 정돈,
성과, 소모 — 300 rows each). Targets contain the correct choice marker and its
full text. The source file hash is
5ff8588998f1cd4a17ddda6a21fe1f50ca519e9fad73fddd8b96a96322368f39.
Dataset page: https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=71949
No other AI Hub dataset, v0.21 mixture, public benchmark question, benchmark
answer, evaluation artifact, log, credential, or .env file is included in
this repository. AI Hub source-data terms remain applicable.
Intended use and limitations
This is an experimental Korean-language fine-tuned model for research and controlled evaluation. It can produce factual or reasoning errors and is not a substitute for professional advice. The HyperCLOVA X acceptable-use restrictions and all applicable laws continue to apply to this derivative model.
License and notices
The full HyperCLOVA X SEED 14B Think Model License Agreement is included in
LICENSE, and the required NAVER attribution is in NOTICE. Redistribution
must comply with that agreement and the AI Hub source-data terms.
Usage
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "jwg0830/HyperCLOVA-X-SEED-Think-14B-sft-71949-3000"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id, dtype=torch.bfloat16, device_map="auto"
)
messages = [{"role": "user", "content": "대한민국의 수도는 어디인가요?"}]
inputs = tokenizer.apply_chat_template(
messages, add_generation_prompt=True, tokenize=True, return_tensors="pt"
).to(model.device)
with torch.inference_mode():
outputs = model.generate(**inputs, max_new_tokens=64, do_sample=False)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))