Evidence: benchmark results vendor-reported
LFM3-VL 2B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
The vision-language member of LFM3: the same 24-layer tri-hybrid backbone with an image encoder, 1.9B parameters in total and a 128K context. Official GGUF, MLX and ONNX builds; llama.cpp has loaded it since 27 October 2026.
LFM3-VL 2B needs about 2 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 1.5 GB | ~317 tok/s | Fits | |
| Q3_K_M 3.41 bpw | 1.7 GB | ~294 tok/s | Fits | |
| Q4_K_M 4.83 bpw | 2.1 GB | ~259 tok/s | Fits | |
| Q5_K_M 5.67 bpw | 2.3 GB | ~243 tok/s | Fits | |
| Q6_K 6.56 bpw | 2.5 GB | ~227 tok/s | Fits | |
| Q8_0 8.50 bpw | 2.9 GB | ~199 tok/s | Fits | |
| F16 16.00 bpw | 4.7 GB | ~135 tok/s | Fits |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. LFM3-VL 2B VRAM calculator →
Get LFM3-VL 2B running
The cheapest catalogued GPU that runs LFM3-VL 2B is the Intel Arc B570 (10 GB).
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run LFM3-VL 2B
No Ollama tag for the vision model yet. Use the official GGUF with llama.cpp, or the MLX builds on Apple Silicon.
Specifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 1.9 Billion
- Context window
- 128,000
- Architecture
- Hybrid: 13 short-conv + 5 sliding-window + 6 global-attention layers, image encoder
- Provider
- Liquid AI
- Licence
- LFM Open License v1.0
- Practical deployment
- Edge device (phone, SBC)
- Specified at
- Q4_K_M
- System RAM
- 8 GB
- Record updated
- 2026-11-03
Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.
Commercial use is licensed only below $10M annual revenue; above that, a separate agreement.
Limits and caveats
- Commercial use is licensed only below US$10M annual revenue.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Other Liquid LFM3 sizes
LFM3-VL 2B — frequently asked questions
How much VRAM does LFM3-VL 2B need?
About 2 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does LFM3-VL 2B run on an RTX 4090 (24 GB)?
Yes. LFM3-VL 2B needs about 2 GB at Q4_K_M, inside a 24 GB card, at an estimated 259 tokens/sec.
How do I run LFM3-VL 2B locally?
No Ollama tag for the vision model yet. Use the official GGUF with llama.cpp, or the MLX builds on Apple Silicon. Running the published tag would send your prompts to a hosted GPU rather than your own machine.
What other sizes does Liquid LFM3 come in?
LFM3 1.7B (2 GB), LFM3-VL 2B (2 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.