This repository provides a merged model based on Qwen/Qwen2.5-7B-Instruct. A LoRA adapter was trained with Unsloth + PEFT, then merged into the base weights.

Training Data

u-10bei/sft_alfworld_trajectory_dataset_v5
u-10bei/dbbench_sft_dataset_react_v4

Training Configuration (summary)

Max sequence length: 4096
Epochs: 1
Learning rate: 1e-05
LoRA: r=64, alpha=128, dropout=0.0
Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Intended Use

This model is tuned for multi-turn agent-style trajectories:

ALFWorld-style household action selection
DBBench-style SQL operation/answer formatting

Notes

Please follow the base model's original license/terms.
The datasets listed above are synthetic and published under their respective licenses.

tomtom5560/qwen2p5-7b-stage1-merged-v4

qwen2.5-7b-alfworld-dbbench-stage1-merged

Training Data

Training Configuration (summary)

Intended Use

Notes

tomtom5560/
qwen2p5-7b-stage1-merged-v4