This adapter was trained using a repaired version of the v5 dataset, ensuring that the model learns from high-quality, syntactically correct structural data (JSON, XML, YAML, TOML, CSV).

Training Objective

This adapter is trained to improve structured output accuracy and format compliance.

Unlike basic fine-tuning, this model has been trained with both the intermediate reasoning (Chain-of-Thought) and the final output, allowing it to "think" about the data structure before generation.

Training Configuration

Base model: Qwen/Qwen3-4B-Instruct-2507
Method: QLoRA (4-bit)
Max sequence length: 2048
Epochs: 3
Learning rate: 1e-04
LoRA: r=64, alpha=64

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = uskma7151/qwen3-4b-v5-3-refined-reasoning"

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

Sources & Terms (IMPORTANT)

Training data: u-10bei/structured_data_with_cot_dataset_512_v5

Dataset License: MIT License. This dataset is used and distributed under the terms of the MIT License. Compliance: Users must comply with the MIT license (including copyright notice) and the base model's original terms of use.