ggml-medium.bin is widely considered the "sweet spot" for local transcription using whisper.cpp
To use this file, a user typically follows a simple but precise ritual: ggml-medium.bin
HIPBLAS success story on AMD graphics · ggml-org whisper.cpp ggml-medium
GGML (now largely superseded by GGUF, but still widely used) is a tensor library for machine learning designed for and running on commodity hardware (CPUs). Created by Georgi Gerganov, the GGML format allows AI models to run on Apple Silicon (M1/M2/M3), Intel CPUs, and even Raspberry Pis by sacrificing a tiny bit of accuracy for massive speed gains. It contains approximately 769 million parameters , quantized
If you encounter ggml-medium.bin , 99% of the time it is converted to GGML format. It contains approximately 769 million parameters , quantized to typically 5-bit or 8-bit integer precision (e.g., q5_0 or q8_0 ).