Name: qwen3-4b-sft-merged-output-lora
Availability: InStock

You can deploy this model and start earning money today!

＜qwen3-4b-sft-merged-output-lora＞

This repository provides a LoRA adapter fine-tuned from Qwen/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. To achieve high reliability, strategic dataset weighting was applied:

Core Structured Data: Selected high-priority datasets were up-sampled by a factor of 2x (double-weighted) to prioritize formatting precision and robustness against complex schemas.
Reasoning Balance: This repetition ensures the model prioritizes these structures while maintaining its general instruction-following capabilities.

Training Configuration

Base model: Qwen/Qwen3-4B-Instruct-2507
Method: QLoRA (4-bit)
Max sequence length: 2048
Epochs: 1
Learning rate: 2e-04
LoRA: r=64, alpha=128

Usage

python

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "your_id/your-repo"

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_v2
u-10bei/structured_data_with_cot_dataset_512_v4
u-10bei/structured_data_with_cot_dataset_512_v5
u-10bei/structured_data_with_cot_dataset_512
u-10bei/structured_data_with_cot_dataset_v2 (Up-sampled 2x)
u-10bei/structured_data_with_cot_dataset (Up-sampled 2x)
daichira/structured-3k-mix-sft (Up-sampled 2x)
daichira/structured-5k-mix-sft (Up-sampled 2x)
daichira/structured-hard-sft-4k (Up-sampled 2x)

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.

Razuka/qwen3-4b-sft-merged-output-lora

Training Objective

Training Configuration

Usage

Sources & Terms (IMPORTANT)

Razuka/
qwen3-4b-sft-merged-output-lora