Went through the SDK docs before asking. On RN/Expo specifically, does Fabric run inside a Bare worklet with IPC back to Hermes, or drop into a native module th

Went through the SDK docs before asking. On RN/Expo specifically, does Fabric run inside a Bare worklet with IPC back to Hermes, or drop into a native module the way llama.rn does via JNI and llama.cpp? Perf and memory footprint would look very different between the two, curious which path you landed on.