File	Quantization	Size
`LFM2.5-Audio-1.5B-JP-F32.gguf`	F32 (language model)	4.4 GB
`LFM2.5-Audio-1.5B-JP-F16.gguf`	F16 (language model)	2.2 GB
`LFM2.5-Audio-1.5B-JP-Q8_0.gguf`	Q8_0 (language model)	1.2 GB
`LFM2.5-Audio-1.5B-JP-Q4_0.gguf`	Q4_0 (language model)	664 MB
`mmproj-LFM2.5-Audio-1.5B-JP-F32.gguf`	F32 (audio encoder / multimodal projector)	695 MB
`mmproj-LFM2.5-Audio-1.5B-JP-F16.gguf`	F16 (audio encoder / multimodal projector)	413 MB
`mmproj-LFM2.5-Audio-1.5B-JP-Q8_0.gguf`	Q8_0 (audio encoder / multimodal projector)	280 MB
`mmproj-LFM2.5-Audio-1.5B-JP-Q4_0.gguf`	Q4_0 (audio encoder / multimodal projector)	210 MB
`vocoder-LFM2.5-Audio-1.5B-JP-F32.gguf`	F32 (vocoder / audio detokenizer)	739 MB
`vocoder-LFM2.5-Audio-1.5B-JP-F16.gguf`	F16 (vocoder / audio detokenizer)	370 MB
`vocoder-LFM2.5-Audio-1.5B-JP-Q8_0.gguf`	Q8_0 (vocoder / audio detokenizer)	197 MB
`vocoder-LFM2.5-Audio-1.5B-JP-Q4_0.gguf`	Q4_0 (vocoder / audio detokenizer)	104 MB
`tokenizer-LFM2.5-Audio-1.5B-JP-F32.gguf`	F32 (audio tokenizer)	268 MB
`tokenizer-LFM2.5-Audio-1.5B-JP-F16.gguf`	F16 (audio tokenizer)	134 MB
`tokenizer-LFM2.5-Audio-1.5B-JP-Q8_0.gguf`	Q8_0 (audio tokenizer)	72 MB
`tokenizer-LFM2.5-Audio-1.5B-JP-Q4_0.gguf`	Q4_0 (audio tokenizer)	46 MB

Runners

runners folder contains pre-built binaries for various architectures:

llama-liquid-audio-cli
llama-liquid-audio-server

🏃 How to run LFM2.5-Audio-JP

CLI

Set env variables.

export CKPT=/path/to/LFM2.5-Audio-1.5B-JP-GGUF
export INPUT_WAV=/path/to/input.wav
export OUTPUT_WAV=/path/to/output.wav

ASR (audio -> text)

./llama-liquid-audio-cli -m $CKPT/LFM2.5-Audio-1.5B-Q4_0.gguf -mm $CKPT/mmproj-LFM2.5-Audio-1.5B-Q4_0.gguf -mv $CKPT/vocoder-LFM2.5-Audio-1.5B-Q4_0.gguf --tts-speaker-file $CKPT/tokenizer-LFM2.5-Audio-1.5B-Q4_0.gguf -sys "Perform ASR in japanese." --audio $INPUT_WAV

TTS (text -> audio)

./llama-liquid-audio-cli -m $CKPT/LFM2.5-Audio-1.5B-Q4_0.gguf -mm $CKPT/mmproj-LFM2.5-Audio-1.5B-Q4_0.gguf -mv $CKPT/vocoder-LFM2.5-Audio-1.5B-Q4_0.gguf --tts-speaker-file $CKPT/tokenizer-LFM2.5-Audio-1.5B-Q4_0.gguf -sys "Perform TTS in japanese." -p "こんにちは、お元気ですか？" --output $OUTPUT_WAV

Interleaved (audio/text -> audio + text)

./llama-liquid-audio-cli -m $CKPT/LFM2.5-Audio-1.5B-Q4_0.gguf -mm $CKPT/mmproj-LFM2.5-Audio-1.5B-Q4_0.gguf -mv $CKPT/vocoder-LFM2.5-Audio-1.5B-Q4_0.gguf --tts-speaker-file $CKPT/tokenizer-LFM2.5-Audio-1.5B-Q4_0.gguf -sys "Respond with interleaved text and audio." --audio $INPUT_WAV --output $OUTPUT_WAV

Server

Start server

export CKPT=/path/to/LFM2.5-Audio-1.5B-JP-GGUF
./llama-liquid-audio-server -m $CKPT/LFM2.5-Audio-1.5B-Q4_0.gguf -mm $CKPT/mmproj-LFM2.5-Audio-1.5B-Q4_0.gguf -mv $CKPT/vocoder-LFM2.5-Audio-1.5B-Q4_0.gguf --tts-speaker-file $CKPT/tokenizer-LFM2.5-Audio-1.5B-Q4_0.gguf

Use liquid_audio_chat.py script to communicate with the server.

uv run liquid_audio_chat.py

Source Code for Runners

Runners are built from https://github.com/ggml-org/llama.cpp/pull/18641.

📬 Contact

Got questions or want to connect? Join our Discord community
If you are interested in custom solutions with edge deployment, please contact our sales team.

License

The code in this repository and associated weights are licensed under the LFM Open License v1.0.

The code for the audio encoder is based on Nvidia NeMo, licensed under Apache 2.0, and the canary-180m-flash checkpoint, licensed under CC-BY 4.0. To simplify dependency resolution, we also ship the Python code of Kyutai Mimi, licensed under the MIT License. We also redistribute weights for Kyutai Mimi, licensed under CC-BY-4.0.

Citation

@article{liquidai2025lfm2,
 title={LFM2 Technical Report},
 author={Liquid AI},
 journal={arXiv preprint arXiv:2511.23404},
 year={2025}
}

LiquidAI/LFM2.5-Audio-1.5B-JP-GGUF

LFM2.5-Audio-1.5B-JP

Available files