This model is a fine-tuned version of Qwen/Qwen3-4B-Instruct using Direct Preference Optimization (DPO) via the Unsloth library. The model has been merged (16-bit), so you can use it directly with the transformers library without additional adapters.

Training Details

Method: DPO (Direct Preference Optimization)
Base Model: Qwen/Qwen3-4B-Instruct-2507
Dataset: u-10bei/dpo-dataset-qwen-cot (Chain-of-Thought)
Framework: Unsloth

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "takenoko888/Qwen3-4B-DPO-Merged"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

prompt = "思考の連鎖を使って、次の問題を解いてください。\nリンゴが3個あります..."
messages = [{"role": "user", "content": prompt}]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to("cuda")

outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))