Chattso-GPT/adv-sft-v2

Qwen3-4B Agent SFT (ALFWorld + DBBench Mixed)

This repository provides a LoRA adapter fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using LoRA + Unsloth.

This repository contains LoRA adapter weights only. The base model must be loaded separately.

Training Objective

This adapter is trained to improve multi-turn agent task performance on both ALFWorld (household tasks) and DBBench (database operations).

Loss is applied to all assistant turns in the multi-turn trajectory, enabling the model to learn environment observation, action selection, tool use, SQL construction, and recovery from errors.

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Method: LoRA (full precision base) + Unsloth
• Datasets: ALFWorld v5 (success only) + DBBench v4 (mixed)
• Max sequence length: 2048
• Epochs: 2
• Learning rate: 2e-6
• Batch size: 2 x grad_accum 4 = effective 8
• LoRA: r=64, alpha=128
• Scheduler: cosine with warmup 10%
• Loss: all assistant turns (multi-turn tool-use pattern)

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "Chattso-GPT/adv-sft-v2"

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/sft_alfworld_trajectory_dataset_v5
• u-10bei/dbbench_sft_dataset_react_v4

Dataset License: MIT License. Compliance: Users must comply with the MIT license and the base model terms of use.