agentbench-qwen3-4b-2stage-reasoning-20260228

A full model fine-tuned from melon1891/agentbench-qwen3-4b-lr5e6-20260224v2 using LoRA + Unsloth, with the adapter merged into 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: melon1891/agentbench-qwen3-4b-lr5e6-20260224v2
• Method: LoRA (merged into base)
• Max sequence length: 8192
• Epochs: 3
• Learning rate: 1e-06
• LoRA: r=16, alpha=32

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("melon1891/agentbench-qwen3-4b-2stage-reasoning-20260228")
tokenizer = AutoTokenizer.from_pretrained("melon1891/agentbench-qwen3-4b-2stage-reasoning-20260228")

Sources & Terms (IMPORTANT)

Training data: melon1891/reasoning-chain-distilled-317

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.