LFM3-Omni-2B needs about 3.2 GB of memory at Q4_K_M (a 1.1 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 7 quants, from 0.8 GB (IQ3_XXS) to 1.8 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 6.4 GB.
LFM3-Omni-2B – a compact omni‑modal model for text, image, video and audio
LFM3-Omni-2B is a 2.17B parameter model that supports text, images, video and audio input and outputs text or speech. It is designed for on‑device use and runs locally in llama.cpp with the LFM3 omni/audio support. The model can process 32K tokens of context and includes a projector for vision and audio.
Good at
- Omni‑modal input and output
- On‑device inference
- Supports 32K token context
Limits
- No reasoning mode in llama.cpp
- Requires projector file for vision/audio
- License type is other
Best for
- Voice‑controlled photo queries
- Transcribing and translating speech
- Chatting with video and audio context
Who it’s for. Developers building edge applications that need a single model for text, vision, audio and speech. Ideal for mobile, laptop or embedded devices where model size and inference speed matter.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ3_XXS | 0.8 GB | 2.9 GB | Fits | Get file → | |
| IQ3_S | 0.9 GB | 3.0 GB | Fits | Get file → | |
| IQ4_XS | 1.1 GB | 3.2 GB | Fits | Get file → | |
| Q4_K_M | 1.1 GB | 3.2 GB | Fits | Get file → | |
| Q5_K_M | 1.3 GB | 3.4 GB | Fits | Get file → | |
| Q6_K | 1.6 GB | 3.7 GB | Fits | Get file → | |
| Q8_0 | 1.8 GB | 4.0 GB | Fits | Get file → |
For 8 GB VRAM, use IQ3_XXS (0.81 GB, 2.9 GB memory) or IQ3_S (0.92 GB, 3 GB memory). For 12 GB VRAM, Q4_K_M (1.13 GB, 3.2 GB memory) or Q5_K_M (1.31 GB, 3.4 GB memory) are suitable. Larger VRAM allows Q6_K (1.59 GB, 3.7 GB memory) or Q8_0 (1.85 GB, 4 GB memory).
What GPU do you need for LFM3-Omni-2B?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is Q8_0 (1.8 GB file, about 4.0 GB in use).
Which LFM3-Omni-2B quant should you choose?
Q4_K_M is the usual starting point for LFM3-Omni-2B: it balances file size and quality well. Here it is a 1.1 GB file.
The Q8_0 file is 1.8 GB, about 1.6 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is IQ3_XXS at 0.8 GB, with the quality trade-offs that come with fewer bits. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about LFM3-Omni-2B
How much VRAM does LFM3-Omni-2B need?
About 3.2 GB at Q4_K_M with an 8K context window. That is the 1.1 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run LFM3-Omni-2B on an 8 GB GPU?
Yes, with Q8_0: the 1.8 GB file needs about 4.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of LFM3-Omni-2B GGUF?
It depends on the quant: IQ3_XXS is 0.8 GB, IQ3_S is 0.9 GB, IQ4_XS is 1.1 GB, Q4_K_M is 1.1 GB, Q5_K_M is 1.3 GB, Q6_K is 1.6 GB and 1 more.
What context length does LFM3-Omni-2B support?
The GGUF header lists a trained context length of 128,000 tokens (125K). The KV cache grows with the context you use: about 1.6 GB at 8K and 6.4 GB at 32K.
Where can I download LFM3-Omni-2B GGUF?
The files are on Hugging Face in the AtomicChat/LFM3-Omni-2B-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does LFM3-Omni-2B use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run LFM3-Omni-2B on 8 GB VRAM?
Yes, the IQ3_XXS or IQ3_S quantizations use about 3 GB of VRAM at 8K context, which fits within 8 GB.
Does the model support vision input?
Yes, it accepts images and video at native resolution, but requires the projector file mmproj-LFM3-Omni-2B-F16.gguf.
Is the model capable of reasoning?
The base model is not a reasoning model; enable reasoning in llama.cpp with the --reasoning off flag.