LLM2025_final_03021100

qwen3-4b-structured-output-lora

This repository provides a LoRA adapter fine-tuned from Qwen3-4B-Instruct-2507 using QLoRA (4-bit, Unsloth).

This repository contains LoRA adapter weights only. The base model must be loaded separately.

Training Objective

This adapter is trained to improve structured output accuracy (JSON / YAML / XML / TOML / CSV).

Loss is applied only to the final assistant output, while intermediate reasoning (Chain-of-Thought) is masked (SFT_MASK_COT=1).

• Output markers: Output:, OUTPUT:, Final:, Answer:, Result:, Response:
• Learn mode: from_marker (loss is computed from the marker token onward, inclusive)

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Method: QLoRA (4-bit)
• Max sequence length: 1024
• Epochs: 1
• Learning rate: 2e-6
• LoRA: r=64, alpha=128
• LoRA dropout: 0.05
• LoRA target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Per-device train batch size: 16
• Per-device eval batch size: 16
• Gradient accumulation steps: 2
• Warmup ratio: 0.1
• Weight decay: 0.05
• Precision: fp16
• Hardware: Google Colab A100

Changes from Baseline Notebook

Training Results

• Train loss: TBD
• Eval loss: TBD
• Total steps: TBD
• Training time: TBD

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "kazfujita/LLM2025_final_03021100"

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_v4

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.