qwen3-4b-agent-full-v3
This repository provides a fully fine-tuned model based on
Qwen/Qwen3-4B-Instruct-2507.
Because this model underwent full parameter fine-tuning, this repository contains the full model weights.
You can load it directly without needing to merge it with the base model.
Training Objective
This model 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: full
- Max sequence length: 4096
- Epochs: 2
- Learning rate: 2e-06
Usage
1from transformers import AutoModelForCausalLM, AutoTokenizer
2import torch
3
4base = "Qwen/Qwen3-4B-Instruct-2507"
5
6tokenizer = AutoTokenizer.from_pretrained(base)
7model = AutoModelForCausalLM.from_pretrained(
8 base,
9 torch_dtype=torch.float16,
10 device_map="auto",
11)
Sources & Terms (IMPORTANT)
Training data: u-10bei/sft_alfworld_trajectory_dataset_v4, u-10bei/sft_alfworld_trajectory_dataset_v5, u-10bei/dbbench_sft_dataset_react_v4, u-10bei/dbbench_sft_dataset_react_v3
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.
