Yejin Korean 3B v2.5 SFT (Step 14K)

A Korean-English bilingual instruction-following model with 3 billion parameters.

Model Details

• Base Model: Yejin Korean 3B v2.5 Base (pre-trained + annealed)
• Parameters: 3B
• Languages: Korean, English
• Context Length: 4096 tokens
• Training: Supervised Fine-Tuning on high-quality Korean instruction data
• SFT Steps: 14,000 (partial training, Loss: 1.58)
• Precision: BF16

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cooler8/yejin-korean-3b-v25-sft-step14k")
model = AutoModelForCausalLM.from_pretrained("cooler8/yejin-korean-3b-v25-sft-step14k", torch_dtype="auto")

prompt = "### 질문:\n대한민국의 수도는 어디인가요?\n\n### 답변:\n"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

License

Apache 2.0