takeofuture/shibatake-SFT_2602071956

qwen3-4b-structured-output-lora-shibatake-v10

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.

• CoT mask: 1
• Learn mode: after_marker
• Output markers: Output:,OUTPUT:,Final:,Answer:,Result:,Response:

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Method: QLoRA (4-bit, Unsloth)
• Max sequence length: 512
• Epochs: 1
• Learning rate: 1e-06
• LoRA: r=64, alpha=128, dropout=0

Usage

pip install -U transformers peft accelerate safetensors

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "takeofuture/shibatake-SFT_2602071956"  # ✅ this repo id

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

Sources & Terms (IMPORTANT)

Training data: daichira/structured-3k-mix-sft Dataset License: MIT License Compliance: Users must comply with the dataset license (including copyright notice) and the base model's original terms of use.

Last updated: 2026-02-07 19:56:47 UTC