Add guardrails to reveal model call

Increase start story temperature and add opening lenses

Fix start story prompt

Fix continue story prompt

Fix LLM output guardrails

Clean up code

Prevent shape judgment model thinking

Revise ollama & hf deployment to llama.cpp

Update ZeroGPU function duration

Fix ZeroGPU decorator

Fix CUDA libs

Add CUDA libs for llama.cpp

Add llama.cpp for shape model

Fix HF model pipeline load

Improve shape card, fix shape card title, fix background

Add shape card for reveal & painting

Add material to shape

Add modal backend for shape reasoning model

ZeroGPU fix attempt

Add is_torch_fx_available fix for loading MiniCPM4.1-8B

Add ZeroGPU support

Initial push