ZYR3 v1 is an agentic coding LoRA adapter on Qwen2.5-Coder-7B-Instruct

The author has released ZYR3 v1, a new agentic coding model available on Hugging Face as a LoRA adapter built on top of Qwen2.5-Coder-7B-Instruct.

• Designed to function as a coding agent rather than a simple completion tool, supporting code generation, debugging, refactoring, and multi-step problem solving.
• Utilizes a workflow of understanding, planning, coding, debugging, and refining to handle real development tasks.
• Available in Safetensors format for use with Transformers, PEFT, vLLM, SGLang, and llama.cpp on modern NVIDIA GPUs.
• A free API is planned to allow developers to test the model via browser or applications without downloading weights.

The release aims to provide a tool that works through problems iteratively, making it more suitable for complex development workflows than standard code generation models.