LLMAdv — Qwen3-4B-Instruct-2507 NF4 Quantized Model

This repository contains a lightweight, 4-bit NF4 quantized version of Qwen3-4B-Instruct-2507, optimized for efficient inference on CPU and Intel GPU environments (e.g., Arc 140T, WSL2).
The model is prepared for competition submission and general-purpose Japanese/English instruction following.

---

Model Details

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Quantization: bitsandbytes 4-bit (NF4)
• Format: model.safetensors
• Tokenizer: Qwen3 tokenizer (included)
• Intended use:
  • Lightweight inference
  • Educational / research use
  • Competition submission
  • Japanese + English instruction following

---

Usage Example

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "rehabiliworld1/LLMAdv"

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

inputs = tokenizer("歩行周期を説明してください。", return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=200)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))