whisper.cpp 1.7.5 tiny.en quantization esp32-s3
whisper.cpp 1.7.5 puts tiny.en q4_0 near 24MB for ESP32-S3
A reader assumes whisper.cpp on a microcontroller is a hack, but 1.7.5’s q4_0 quantization plus ggml’s mmap means the 24 MB model never has to fit in RAM.