Only with CPU offload

LFM2.5 8B-A1B at Q4_K_M needs 5.5 GB but only 5.3 GB is addressable, so about 3% of the layers would stream from system RAM at roughly 60 GB/s. Expect around 81 tokens per second — usable for batch work, painful for chat.

97% on GPU 8K context Q4_K_M · 4.8 GB LFM Open License v1.0 Released 28 May 2026

An 8B MoE with ~1.5B active, aimed at laptops without a GPU. Licence is permissive below $10M revenue.

Hardware requirements What hardware do I need to run LFM2.5 8B-A1B? →

At Q4_K_M with an 8K context: a 8 GB GPU (RTX 4060 Laptop) or a 16 GB Mac (M4). Minimum, recommended, high-quality and long-context tiers, with all 114 devices checked at every quantisation.

Fits instead: LFM2.5 2.6B (2.2 GB)

The VRAM budget

weights 4.8 GB
Weights 4.8 GB KV cache @ 8K 0.09 GB Runtime overhead 0.6 GB Over budget 0.2 GB past 5.3 GB

Quantisation ladder

QuantWeightsTotal @ 8KMax contextTok/sQualityFit
F16 15.8 GB 16.5 GB — ~6.6 Reference 11.2 GB over
Q8_0 8.4 GB 9.1 GB — ~17 −0.1% ppl 3.8 GB over
Q6_K 6.5 GB 7.2 GB — ~29 −0.4% ppl 1.9 GB over
Q5_K_M 5.6 GB 6.3 GB — ~44 −0.8% ppl 1.0 GB over
Q4_K_M 4.8 GB 5.5 GB — ~81 −1.9% ppl 0.2 GB over
Q3_K_M 3.9 GB 4.5 GB 72K 115 −5.4% ppl Long context
Q2_K 3.3 GB 4.0 GB 119K 134 −15% ppl Long context

Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 6 of its 24 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.

How to run it

terminal
$ llama-server \
    -hf LiquidAI/LFM2.5-8B-A1B:Q4_K_M \
    -c 8192 -ngl 23

The engine underneath most of the others. Every knob is exposed. More on llama.cpp.

01Download is 4.8 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03If it falls back to CPU silently, drop the context first, then step down a quantisation.
See all models for this rig Compare with another model

Related models on a GeForce GTX 1660 Super

ModelTotalTok/s est.
LFM2.5 2.6B
2.7B · same family
Q4_K_M · 2.2 GB 134 Runs great
Fara 7B
8.29B · similar size
Q4_K_M · 5.7 GB ~31 CPU offload
Qwen3 8B
8.19B · similar size
Q4_K_M · 6.3 GB ~22 CPU offload
Granite 4.1 8B
8.79B · similar size
Q4_K_M · 6.8 GB ~17 CPU offload
Ministral 3 8B
8.92B · similar size
Q4_K_M · 6.7 GB ~18 CPU offload
Llama 3.1 8B Instruct
8.03B · similar size
Q4_K_M · 6.1 GB ~25 CPU offload
Gemma 4 E4B
8.0B · similar size
Q4_K_M · 5.2 GB 45 Tight fit

Every model against the GeForce GTX 1660 Super →

LFM2.5 8B-A1B on other hardware

DeviceTotalTok/s est.
GeForce RTX 4060 Laptop
7.0 GB usable · 256 GB/s
5.5 GB 71 Runs great
GeForce RTX 3060 12 GB
10.6 GB usable · 360 GB/s
5.5 GB 99 Runs great
M4 · 16 GB
10.7 GB usable · 120 GB/s
5.5 GB 37 Runs great
GeForce RTX 4090
22.4 GB usable · 1008 GB/s
5.5 GB 278 Runs great
Radeon RX 7900 XTX
22.4 GB usable · 960 GB/s
5.5 GB 265 Runs great
M3 Max · 64 GB
48.0 GB usable · 400 GB/s
5.5 GB 124 Runs great

All 114 devices, by tier →