tropico0313/my-lora-test

qwen3-4b-structured-output-lora

This repository provides a LoRA adapter fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using QLoRA (4-bit) with 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 (assistant-only loss).

Chain-of-Thought masking: Enabled
Learning mode: after_marker

Data Preprocessing

Rule-based normalization was applied before training:

• Extracting content after output markers
• Removing code fences (json / yaml / xml / toml)
• Removing leading boilerplate and trailing notes
• Recursive JSON exact-match deduplication

Dedupe enabled: Yes

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Method: QLoRA (4-bit) + Unsloth
• Max sequence length: 1024
• Epochs: 1
• Learning rate: 3e-05
• Warmup ratio: 0.06
• Weight decay: 0.02
• LoRA: r=48, alpha=96, dropout=0.06
• Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "tropico0313/my-lora-test"

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 dataset: u-10bei/structured_data_with_cot_dataset_512_v2
Dataset License: MIT License.
Users must comply with the MIT license (including copyright notice)
and the base model's original terms of use.