---
base_model: Qwen/Qwen3-4B-Instruct-2507
datasets:
- u-10bei/sft_alfworld_trajectory_dataset_v2
- u-10bei/sft_alfworld_trajectory_dataset_v3
- u-10bei/sft_alfworld_trajectory_dataset_v7
- u-10bei/sft_alfworld_trajectory_dataset_v5
- u-10bei/dbbench_sft_dataset_react
- u-10bei/dbbench_sft_dataset_react_v2
- u-10bei/dbbench_sft_dataset_react_v3
- u-10bei/dbbench_sft_dataset_react_v7
language:
- en
license: apache-2.0
library_name: peft
pipeline_tag: text-generation
tags:
- lora
- agent
- alfworld
- dbbench
- agentbench
---

# Qwen3-4B Agent SFT v7 (All Datasets + Optimized)

LoRA adapter fine-tuned from **Qwen/Qwen3-4B-Instruct-2507** using LoRA + Unsloth.
This repository contains LoRA adapter weights only.

## Training Configuration
- Base model: Qwen/Qwen3-4B-Instruct-2507
- Method: LoRA (full precision) + Unsloth
- Datasets: ALFWorld v2-v5 (deduplicated, EEF) + DBBench v1-v7 (deduplicated, 2x upsampled)
- Max sequence length: 4096
- Epochs: 1
- Learning rate: 2e-6
- LoRA: r=64, alpha=128
- Scheduler: cosine with warmup 10%

## Sources & Terms
Dataset License: MIT License.