Fix: split hardcoded/GPU paths, duration=30, no GPU waste on identity

Fix: hardcode identity responses, remove primer complexity

Improve: identity priming, tighter params, GPU cleanup, curated examples

Add error surfacing + device logging for ZeroGPU debugging

Fix ZeroGPU: explicit model.to(device) inside @spaces.GPU(duration=60)

Upgrade to ZeroGPU (A10G): bfloat16, @spaces.GPU decorator, temp=0.6

Tune: lower temperature to 0.4 for CPU inference quality

Fix: pin gradio==5.29.0 (Python 3.13 pydub fix)

Fix: gradio 5.x for Python 3.13 compatibility

Initial Gruk Space: Gradio app with anti-looping inference

initial commit