Overview

This model, y-ohtani/qwen3-4b-agent-sft-true, is a full fine-tuned version of the Qwen3-4B-Instruct-2507 base model, developed by y-ohtani. Unlike LoRA adapters, this model has undergone comprehensive training to adapt its entire parameter set. It leverages the Open-AgentRL framework for its fine-tuning process, specifically designed for agentic capabilities.

Key Capabilities

• Agentic SFT: Trained with multi-turn agentic Supervised Fine-Tuning (SFT) using real End-to-End agentic trajectories.
• Extended Context: Supports a maximum sequence length of 32,768 tokens, enabling processing of lengthy and complex conversations.
• Specialized Dataset: Fine-tuned on the Gen-Verse/Open-AgentRL-SFT-3K dataset, comprising 3,000 multi-turn conversations tailored for agent behavior.
• Full Fine-tuning: Utilizes FSDP with bfloat16 for robust and efficient full model fine-tuning.

Ideal Use Cases

This model is particularly well-suited for applications requiring:

• Complex Multi-turn Dialogues: Excels in scenarios where an AI agent needs to maintain context and perform multi-step reasoning over several turns.
• Agent-based Systems: Designed for integration into systems that require an AI to act as an agent, performing tasks or solving problems through interactive conversation.
• Instruction Following: Enhanced ability to follow intricate instructions and generate coherent, contextually relevant responses in agentic workflows.