ishikawakazuhiko/qwen3-4b-sft-t4-mix-r60d40-latest

qwen3-4b-agent-trajectory-lora-t4-mixsearch

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 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, and recovery from errors.

Training Configuration

• —Base model: Qwen/Qwen3-4B-Instruct-2507
• —Method: LoRA
• —Max sequence length: 1024
• —Epochs: 1
• —Learning rate: 3e-05
• —LoRA: r=64, alpha=128
• —Mixed training: True
• —Mix ratio (alfworld/dbbench): 0.6/0.4

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = os.environ.get("SFT_HF_UPLOAD_REPO_ID", "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/sftalfworldtrajectory_dataset
• —u-10bei/sftalfworldtrajectorydatasetv2
• —u-10bei/sftalfworldtrajectorydatasetv3
• —u-10bei/sftalfworldtrajectorydatasetv4
• —u-10bei/sftalfworldtrajectorydatasetv5
• —u-10bei/dbbenchsftdataset_react
• —u-10bei/dbbenchsftdatasetreactv2
• —u-10bei/dbbenchsftdatasetreactv3
• —u-10bei/dbbenchsftdatasetreactv4

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 original terms of use.