Model reference · open weights
LFM3-Audio is an open-weight language model from LiquidAI. LFM3-Audio-2B (BF16) weighs 4.0 GB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the LiquidAI/LFM3-Audio-2B-GGUF model card, 2026-10-29. The estimate below is for another build of the family.
What it is
| Released by | LiquidAI |
|---|---|
| Released | 2026-10-27 |
| VRAM | 4.0 GB for the weights |
What it runs on
How much memory each request adds isn't estimated yet for this architecture. The weights need at least the cards below, plus room for the context.
| Card | Weights alone |
|---|---|
| RTX 3060 12 GB | fits |
| RTX 4060 Ti 16 GB | fits |
| RTX 3090 24 GB | fits |
| RTX 4090 24 GB | fits |
| RTX 5090 32 GB | fits |
| L40S 48 GB | fits |
| A100 80 GB | fits |
| H100 80 GB | fits |
| RTX PRO 6000 Blackwell 96 GB | fits |
| DGX Spark (GB10) 128 GB unified | fits |
| H200 141 GB | fits |
| B200 180 GB | fits |
Builds
| Build | Params | Precision | Weights | Smallest setup |
|---|---|---|---|---|
| LFM3-Audio-2B-GGUF ↗ 5 builds: Q4_0 1.0 GB … F16 3.5 GB |
GGUF | 1.1 GB | RTX 3060 12 GB | |
| LFM3-Audio-2B (above) ↗ | 2.0B | BF16 | 4.0 GB | RTX 3060 12 GB |
From the model card
src="https://huggingface.co/LiquidAI/LFM3-1.7B/resolve/main/assets/liquid-ai-banner.webp" alt="Liquid AI" style="width: 100%; max-width: 100%; height: auto; display: block; margin: 0;" />
GGUF version of LFM3-Audio-2B for local inference on CPUs and GPUs.
LFM3-Audio-2B is a compact audio model that runs locally on laptops and edge devices. It takes text or audio as input and responds with text, speech, or both.
Find architecture and usage details in the original model card.
[!NOTE] Supported in llama.cpp b11960 (v0.8.0) and newer, including real-time voice conversations (with tool calls) on the
/v1/realtimeWebSocket API ofllama-server.
Add a quantization tag to pick a file. With -hf, llama.cpp also downloads the audio projector.
Speech recognition: use the system prompt Perform ASR in ., or Perform ASR. to let the model infer the language.
llama-mtmd-cli -hf LiquidAI/LFM3-Audio-2B-GGUF:Q4_K_M \
--audio speech.wav -sys "Perform ASR in English." -p "" --temp 0
Speech synthesis: --tts-lang takes a language code (default en) and --tts-voice is female (default) or male. The audio sampling parameters of the original model card are stored in the GGUF.
llama-tts -m LFM3-Audio-2B-Q4_K_M.gguf -mm mmproj-LFM3-Audio-2B-Q8_0.gguf \
-p "Hello world" --tts-voice male -o out.wav
More options in the llama.cpp guide.
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.
How it works