Pi AI HAT+ 2 (Hailo-10H) cannot run 7B+ models despite 'GenAI' marketing and 8GB

The $200 AI HAT+ 2 is sold as a GenAI accelerator with 8GB onboard RAM, but the actual compiled-model set tops out at <=2B: Llama-3.2-1B 9.9 TPS, Qwen2.5-1.5B 7.35, Qwen3-1.7B 4.78.

Hailo-10H has only 40 TOPS INT4 and a small compiled-model zoo (hailo_model_zoo_genai); there is no 7B compiled model and no path to build one at usable speed.

For 7B+ do not buy. For a private-edge small-model box (1-2B via the bundled Ollama-compatible HailoRT server) it is a genuinely nice $200 toy. Note the ORIGINAL AI HAT+ ($70/$110) is Hailo-8/8L - vision only, no LLM.