Llama-3-Open-Ko-8B-Instruct-tc1-sft

Llama-3-Open-Ko-8B-Instruct-tc1-sft

This model is a supervised fine-tuned (LoRA merged) version of beomi/Llama-3-Open-Ko-8B-Instruct-preview,
trained for Korean instruction-following and customer-support style responses.

---

🔍 Model Summary

• Base Model: beomi/Llama-3-Open-Ko-8B-Instruct-preview (Llama 3 architecture)
• Type: Causal Language Model (decoder-only)
• Parameters: ~8B
• Training Method: LoRA-based Supervised Fine-tuning (SFT)
• Merged: Yes (LoRA adapter merged into full-precision weights)
• Input format: Korean Q&A, instruction tuning format
• Use Cases: Customer support chatbot, Korean instruction following, general Q&A

---

📦 Included Files

Transformers-compatible weights

• config.json
• generation_config.json
• model.safetensors.index.json
• model-00001-of-00007.safetensors ~ model-00007-of-00007.safetensors
• tokenizer.json
• tokenizer_config.json
• special_tokens_map.json

llama.cpp / Ollama (GGUF) format

• model-f16.gguf
• Modelfile
• chat_template.jinja

This repository provides both HF and GGUF formats, enabling easy use across multiple inference environments.

---

🧪 Usage (Transformers)

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "Nepirity/Llama-3-Open-Ko-8B-Instruct-tc1-sft"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

prompt = "안녕하세요? 어떤 서비스를 도와드릴까요?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))