Fix: use 4-bit NF4 quantization to reduce VRAM/CPU-RAM usage

Fix: use 8-bit quantization to fit 7B model in 14GB T4 VRAM

Fix: use 8-bit quantization to fit 7B model in 14GB T4 VRAM

Fix: add demo.queue() to handle long inference in browser

Fix: remove @spaces.GPU decorator (not needed on T4 hardware)

Fix: remove @spaces.GPU decorator (not needed on T4 hardware)

Fix: load model at startup for T4 GPU (not lazy)

Fix: remove image.submit() not available in Gradio 5

Add README.md

Add requirements.txt

Add app.py

initial commit