The $8 AI Machine: 28.9M-Parameter Model Runs on ESP32-S3 Microcontroller

Developer slvDev has successfully run a 28.9-million-parameter language model on an ESP32-S3 microcontroller costing only $8, completely offline at ~9.5 tokens/s with minimal power draw. By memory-mapping most parameters into flash via Google's Per-Layer Embeddings technique, the model fits within the chip's constraints, generating coherent short stories. This marks a breakthrough in ultra-low-cost local AI.

• 本地AI
• 低成本AI
• 模型压缩
• ESP32-S3
• TinyStories
• 嵌入式
• Flash映射