Single-file CMF build for local inference (Rust; Vulkan, Metal, CPU)
#25
by infosave - opened
We packaged LFM3-Audio-2B as single CMF files with the LiquidCodec inside, running with cortiq, a Rust inference engine with no Python: speech recognition, speech synthesis in 19 languages and interleaved voice chat, on Vulkan/DX12, Metal or the CPU.
Files: https://huggingface.co/infosave/LFM3-Audio-2B-cmf — q8_2f (2.45 GB) and q4tp (1.62 GB). Both transcribe the test clips word for word as Transformers does; an f16 conversion reproduces Transformers' speech frames exactly.
7 s of speech synthesizes in 3.2 s on an Apple M4 and 5.6 s on an NVIDIA A40.
cargo install cortiq-cli
cortiq run lfm3-audio-2b-q8_2f.cmf --system "Perform ASR in English." --audio clip.wav --prompt ""
cortiq run lfm3-audio-2b-q8_2f.cmf --system "Perform TTS in American English using a female voice." \
--prompt "The quick brown fox jumps over the lazy dog." --speech-out tts.wav
(cortiq 0.8.50)