Skip to content

LFM3 1.7B VRAM requirements

How much VRAM LFM3 1.7B needs
4-bit weights: 1.0 GBWith 8K context: 1.8 GBWith 32K context: 2.1 GB8-bit, 8K context: 2.6 GB
Parameters
1.7B
Longest context
256K
Released
Oct 2026
License
Lfm1.0
Strongest at
Not rated

Its licence (Lfm1.0) is not a standard open-source one: check the terms before commercial use.

How much VRAM LFM3 1.7B needs

Memory LFM3 1.7B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit3.2 GB4.1 GB4.4 GB5.5 GB
8-bit1.7 GB2.6 GB2.8 GB4.0 GB
6-bit1.3 GB2.2 GB2.5 GB3.6 GB
5-bit1.1 GB2.0 GB2.3 GB3.4 GB
4-bit1.0 GB1.8 GB2.1 GB3.2 GB
3-bit806 MB1.6 GB1.9 GB3.0 GB
2-bit618 MB1.5 GB1.7 GB2.9 GB
1.6-bit340 MB1.2 GB1.5 GB2.6 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run LFM3 1.7B

The best way LFM3 1.7B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit256K96 tok/s0.2 sPDF: LFM3 1.7B on RTX 4060 report
RTX 50608-bit256K152 tok/s0.2 sPDF: LFM3 1.7B on RTX 5060 report
RTX 30708-bit256K152 tok/s0.2 sPDF: LFM3 1.7B on RTX 3070 report
RTX 3080 10 GB8-bit256K240 tok/s0.1 sPDF: LFM3 1.7B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit256K125 tok/s0.3 sPDF: LFM3 1.7B on RTX 3060 12 GB report
Arc B5808-bit256K105 tok/s0.2 sPDF: LFM3 1.7B on Arc B580 report
RTX 40708-bit256K169 tok/s0.1 sPDF: LFM3 1.7B on RTX 4070 report
RTX 4070 Super8-bit256K169 tok/s0.1 sPDF: LFM3 1.7B on RTX 4070 Super report
RTX 50708-bit256K216 tok/s0.1 sPDF: LFM3 1.7B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit256K102 tok/s0.1 sPDF: LFM3 1.7B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit256K152 tok/s0.1 sPDF: LFM3 1.7B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit256K125 tok/s0.2 sPDF: LFM3 1.7B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit256K167 tok/s0.1 sPDF: LFM3 1.7B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit256K216 tok/s0.1 sPDF: LFM3 1.7B on RTX 4070 Ti Super report
RTX 4080 Super8-bit256K233 tok/s0.1 sPDF: LFM3 1.7B on RTX 4080 Super report
RTX 5070 Ti8-bit256K274 tok/s0.1 sPDF: LFM3 1.7B on RTX 5070 Ti report
RTX 50808-bit256K290 tok/s0.1 sPDF: LFM3 1.7B on RTX 5080 report
Radeon RX 7900 XTX8-bit256K228 tok/s0.1 sPDF: LFM3 1.7B on Radeon RX 7900 XTX report
RTX 30908-bit256K284 tok/s0.1 sPDF: LFM3 1.7B on RTX 3090 report
RTX 40908-bit256K301 tok/s0.1 sPDF: LFM3 1.7B on RTX 4090 report
Radeon AI PRO R97008-bit256K167 tok/s0.1 sPDF: LFM3 1.7B on Radeon AI PRO R9700 report
RTX 50908-bit256K458 tok/s0.1 sPDF: LFM3 1.7B on RTX 5090 report
2× RTX 30908-bit256K284 tok/s0.1 sPDF: LFM3 1.7B on 2× RTX 3090 report
2× RTX 40908-bit256K301 tok/s0.1 sPDF: LFM3 1.7B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K374 tok/s0.1 sPDF: LFM3 1.7B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K290 tok/s0.1 sPDF: LFM3 1.7B on RTX 6000 Ada report
L40S8-bit256K266 tok/s0.1 sPDF: LFM3 1.7B on L40S report
2× RTX 50908-bit256K458 tok/s0.1 sPDF: LFM3 1.7B on 2× RTX 5090 report
4× RTX 30908-bit256K284 tok/s0.1 sPDF: LFM3 1.7B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K458 tok/s0.1 sPDF: LFM3 1.7B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit256K49 tok/s2.1 sPDF: LFM3 1.7B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit256K49 tok/s2.1 sPDF: LFM3 1.7B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit256K49 tok/s2.1 sPDF: LFM3 1.7B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit256K99 tok/s1.1 sPDF: LFM3 1.7B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit256K99 tok/s1.1 sPDF: LFM3 1.7B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K99 tok/s1.1 sPDF: LFM3 1.7B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K156 tok/s0.5 sPDF: LFM3 1.7B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K169 tok/s0.2 sPDF: LFM3 1.7B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K186 tok/s0.3 sPDF: LFM3 1.7B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K186 tok/s0.3 sPDF: LFM3 1.7B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K186 tok/s0.3 sPDF: LFM3 1.7B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K234 tok/s0.1 sPDF: LFM3 1.7B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K76 tok/s0.3 sPDF: LFM3 1.7B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K76 tok/s0.3 sPDF: LFM3 1.7B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K106 tok/s0.1 sPDF: LFM3 1.7B on RTX Spark report
DGX Spark8-bit256K97 tok/s0.1 sPDF: LFM3 1.7B on DGX Spark report
Dual DGX Spark8-bit256K97 tok/s0.1 sPDF: LFM3 1.7B on Dual DGX Spark report
DGX Station GB3008-bit256K918 tok/s0.1 sPDF: LFM3 1.7B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K458 tok/s0.1 sPDF: LFM3 1.7B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K458 tok/s0.1 sPDF: LFM3 1.7B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K458 tok/s0.1 sPDF: LFM3 1.7B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K498 tok/s0.1 sPDF: LFM3 1.7B on A100 80 GB SXM report
H100 SXM8-bit256K665 tok/s0.1 sPDF: LFM3 1.7B on H100 SXM report
4× H200 NVL8-bit256K789 tok/s0.1 sPDF: LFM3 1.7B on 4× H200 NVL report
8× H200 NVL8-bit256K789 tok/s0.1 sPDF: LFM3 1.7B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf LiquidAI/LFM3-1.7B-GGUF
vLLM
vllm serve LiquidAI/LFM3-1.7B
MLX (Mac)
mlx_lm.chat --model mlx-community/LFM3-1.7B-OptiQ-4bit

Questions

How much VRAM does LFM3 1.7B need?

LFM3 1.7B needs 1.0 GB for its weights at 4-bit, 1.8 GB with an 8K context and 2.1 GB with 32K. At 8-bit, 2.6 GB.

What is the minimum VRAM for LFM3 1.7B?

About 1.6 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is LFM3 1.7B?

1.7B parameters: 3.2 GB of weights at 16-bit, 1.0 GB at 4-bit.

Can LFM3 1.7B run on an RTX 4090?

Yes. At 8-bit it holds 256K of context and writes about 301 tokens a second.

More models