Google Coral Edge TPU is INT8 vision-only - no LLM path

No LLM serving path, no transformer-decode support, no int4/GGUF story exists for the Edge TPU.

The Coral stack is LiteRT/EdgeTPU-compiled CNNs (classification, detection, pose, segmentation) at ~4 TOPS INT8 - it cannot hold a 7B weight set, let alone stream one.

None for LLMs. Use it for camera/detection projects; buy a used GPU for language models.

Used units trade at $50-130 and get bundled 'free' with other gear, which is why they keep surfacing in homelab carts.