You can squeeze a lot more performance out with a newer framework and a model tailored for your GPU and task.
I'd recommend:
Kobold.cpp rocm, follow the quick-install guide here: https://github.com/YellowRoseCx/koboldcpp-rocm/?tab=readme-ov-file#quick-linux-install
Download this quantization, which fits in your VRAM pool nicely and is specifically tuned for coding and planning, select it in kobold.cpp: https://huggingface.co/mradermacher/Qwen3-14B-Esper3-i1-GGUF/blob/main/Qwen3-14B-Esper3.i1-IQ4_NL.gguf
Use the "corporate" UI in kobold.cpp in your browser. If that doesn't work well, kobold.cpp also works as a generic OpenAI endpoint, which you can access from pretty much any app, like https://openwebui.com/