Qwen3-4B Agent SFT for ALFWorld & DBBench (2c21 - Nothing Happens Fix)
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 (full precision base)
- Max sequence length: 2048
- Epochs: 2
- Learning rate: 5e-06
- LoRA: r=128, alpha=128
Key Changes in 2c21
Based on 2c14 (best score: ALF 74%, DB 51.8%, Score 4.8386)
Only one change from 2c14:
- Added 1 "Nothing happens" response pattern (no upsampling)
- When the agent receives "Nothing happens", it should verify status instead of assuming completion
All other settings maintained from 2c14:
- ALF samples: 8 × 85 = 680
- DB samples: 5 × 55 = 275
- DBBench data: 5% sampled
- LR: 5.5e-6
- Epochs: 2
- WARMUP_RATIO: 0.1
Usage
1from transformers import AutoModelForCausalLM, AutoTokenizer
2from peft import PeftModel
3import torch
4
5base = "Qwen/Qwen3-4B-Instruct-2507"
6adapter = "TToyo2511/ttoyo_advance_2c21" #★TTT20260301 2c21版
7
8tokenizer = AutoTokenizer.from_pretrained(base)
9model = AutoModelForCausalLM.from_pretrained(
10 base,
11 torch_dtype=torch.float16,
12 device_map="auto",
13)
14model = PeftModel.from_pretrained(model, adapter)
Sources & Terms (IMPORTANT)
Training data:
- u-10bei/sft_alfworld_trajectory_dataset_v5
- u-10bei/dbbench_sft_dataset_react_v4 (5% sampled)
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's original terms of use.
