At a guess, based on model size, just about anything, even a Raspberry Pi 4 etc. As comparison the GGML port of Whisper (OpenAI's equivalent), runs in the brow

At a guess, based on model size, just about anything, even a Raspberry Pi 4 etc. As comparison the GGML port of Whisper (OpenAI's equivalent), runs in the browser, via WASM: https://whisper.ggerganov.com/ https://whisper.ggerganov.com/