yoei/qwen3-4b-agentbench-merged02

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 18, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The yoei/qwen3-4b-agentbench-merged02 is a 4 billion parameter LoRA adapter, fine-tuned from Qwen/Qwen3-4B-Instruct-2507, specifically designed to enhance multi-turn agent task performance. It excels in complex environments like ALFWorld (household tasks) and DBBench (database operations) by learning from full multi-turn trajectories, including environment observation, action selection, tool use, and error recovery. This adapter is optimized for developing intelligent agents capable of sequential decision-making and interaction.

Loading preview...

Overview

yoei/qwen3-4b-agentbench-merged02 is a LoRA adapter fine-tuned from the Qwen/Qwen3-4B-Instruct-2507 base model, utilizing LoRA + Unsloth for efficient training. This repository provides only the adapter weights, requiring the base model to be loaded separately.

Key Capabilities

  • Enhanced Multi-Turn Agent Performance: Specifically trained to improve performance in complex, multi-turn agent tasks.
  • Task Domains: Optimized for household tasks (ALFWorld) and database operations (DBBench).
  • Trajectory Learning: Learns from full multi-turn trajectories, including environment observation, action selection, tool use, and error recovery, by applying loss to all assistant turns.

Good For

  • Developing intelligent agents that require sequential decision-making.
  • Applications involving tool use and interaction with dynamic environments.
  • Research and development in agentic AI and reinforcement learning from human feedback (RLHF) for complex tasks.
  • Users looking to leverage a 4B parameter model for efficient agentic reasoning.