qwen3-4b-structured-output-lora-comp-2026feb-run08

Model Card for structeval RUN08

This adapter was trained for the StructEval competition using supervised fine-tuning (SFT).

Base model

unsloth/qwen3-4b-instruct-2507-unsloth-bnb-4bit

Training configuration

RUN_ID: structeval_20260301_024851_32a2ac1e

Learning rate: 2.6e-6
Epochs: 2
Max seq length: 1536
Batch size: 2
Gradient accumulation: 4

LoRA: r = 64
alpha = 128
dropout = 0.0

Upsampling: {"toml": 3.0}

Validation loss: 0.9484043121337891

Purpose

This LoRA adapter was trained to improve structured output generation (JSON / YAML / TOML / XML / CSV) for the StructEval competition.

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = "unsloth/qwen3-4b-instruct-2507-unsloth-bnb-4bit"
adapter = "yamayos/qwen3-4b-structured-output-lora-comp-2026feb-run08"

model = AutoModelForCausalLM.from_pretrained(base)
model = PeftModel.from_pretrained(model, adapter)

tokenizer = AutoTokenizer.from_pretrained(base)