Fix: barebones Gradio app to test core functionality

Fix: use flan-t5-small with text_generation (no token needed)

Fix: use InferenceClient without token requirement

Simplify: use gr.load with zephyr-7b-beta

Simplify: use zephyr-7b-beta with ChatInterface

Fix: Use microsoft/Phi-3-mini-4k-instruct instead of non-existent model

Deploy AURA Chat Space