set hardware=cpu-basic (proxy doesnt need gpu)

redeploy: simple gradio proxy

switch to thin HF Router proxy (no model loading)

switch to Qwen2.5-Coder-7B-Instruct (drop AWQ)

fix: gradio 5.x for HfFolder import compat

init: Qwen2.5-Coder-32B-Instruct-AWQ via FastAPI/Gradio

initial commit