qwen3-4b-structeval-sft-v1

qwen3-4b-structeval-sft-30g70c-v2

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. (For datasets with intermediate reasoning, training targets are restricted to the final output.)

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Method: QLoRA (4-bit)
• Max sequence length: 1024
• Epochs: 1
• Learning rate: 2e-05
• LoRA: r=64, alpha=128

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "Termaln/qwen3-4b-structeval-sft-v1"

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 datasets:
- u-10bei/structured_data_with_cot_dataset_v2
- daichira/structured-3k-mix-sft
- daichira/structured-5k-mix-sft

Dataset licenses (verify on each dataset card):
- u-10bei/structured_data_with_cot_dataset_v2: (check dataset card on HF Hub)
- daichira/structured-3k-mix-sft: CC-BY-4.0
- daichira/structured-5k-mix-sft: CC-BY-4.0

Base model license: Apache-2.0 (see base model repository)
Adapter repo license (this repo): apache-2.0

Compliance: Users must comply with each dataset's license conditions and the base model's original terms of use.