How much VRAM does LFM2.5 8B-A1B need?

At Q4_K_M the weights are 4.8 GB; with an 8K context KV cache and runtime overhead it needs 5.5 GB, so a 8 GB card runs it with headroom.

Can a GeForce RTX 4070 run LFM2.5 8B-A1B?

Yes — Q6_K fits with headroom at 8K context, about 102 tokens per second estimated.

Can a GeForce RTX 4090 run LFM2.5 8B-A1B?

Yes — Q8_0 fits with headroom at 8K context, about 158 tokens per second estimated.

Can a GeForce RTX 3060 12 GB run LFM2.5 8B-A1B?

Yes — Q6_K fits with headroom at 8K context, about 73 tokens per second estimated.

Can a M3 Max · 64 GB run LFM2.5 8B-A1B?

Yes — Q8_0 fits with headroom at 8K context, about 70 tokens per second estimated.

Hardware requirements

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

8.47B MoE · 1.5B active 125K context LFM Open License v1.0

At Q4_K_M the weights are 4.8 GB; with an 8K KV cache and runtime overhead that is 5.5 GB, so it wants a 8 GB card to run with headroom — in practice 8 GB VRAM: GeForce RTX 3070 Ti and 7 others. Every figure below is computed from the model’s published shape against each device’s usable memory; tokens per second are estimates. The method.

Minimum practical 8 GB VRAM

GeForce RTX 3070 Ti — Q5_K_M, 6.3 GB of 7.0 GB, ~143 tok/s, up to 68K context.

Also: Arc A750, GeForce RTX 5060, GeForce RTX 3070, GeForce RTX 3060 Ti, GeForce RTX 4060 Ti +2 more

the smallest card that holds it at all, at Q3_K_M — the lowest quantisation worth running

Recommended 8 GB VRAM

GeForce RTX 3070 Ti — Q4_K_M, 5.5 GB of 7.0 GB, ~168 tok/s, up to 125K context.

Also: Arc A750, GeForce RTX 5060, GeForce RTX 3070, GeForce RTX 3060 Ti, GeForce RTX 4060 Ti +2 more

runs the recommended quantisation with headroom at 8K context

High quality 16 GB VRAM

GeForce RTX 5080 — Q8_0, 9.1 GB of 14.4 GB, ~151 tok/s, up to 125K context.

Also: GeForce RTX 5070 Ti, GeForce RTX 4080 Super, GeForce RTX 4080, GeForce RTX 4070 Ti Super, Radeon RX 9070 XT +12 more

near-lossless Q8_0 with headroom

Long context 10 GB VRAM

GeForce RTX 3080 — Q4_K_M, 6.8 GB of 8.8 GB, ~210 tok/s at 125K.

Also: Arc B570

128K tokens (or the model’s maximum) at the recommended quantisation

Apple Silicon 16 GB unified

M1 · 16 GB — Q4_K_M, 5.5 GB of 10.7 GB, ~21 tok/s, up to 125K context.

smallest Mac that runs the recommended quantisation with headroom

CPU only 32 GB RAM

CPU only · DDR4 dual-channel — Q8_0, 9.1 GB of 25.6 GB, ~6.4 tok/s, up to 125K context.

no GPU — weights stream from system RAM

How much VRAM at each quantisation

Quant	Weights	+KV 8K	+KV 32K	+KV 128K	Total @ 8K	Needs	Quality
F16	15.8 GB	0.09 GB	0.38 GB	1.5 GB	16.5 GB	24 GB card	Reference
Q8_0	8.4 GB	0.09 GB	0.38 GB	1.5 GB	9.1 GB	16 GB card	−0.1% ppl
Q6_K	6.5 GB	0.09 GB	0.38 GB	1.5 GB	7.2 GB	10 GB card	−0.4% ppl
Q5_K_M	5.6 GB	0.09 GB	0.38 GB	1.5 GB	6.3 GB	10 GB card	−0.8% ppl
Q4_K_M	4.8 GB	0.09 GB	0.38 GB	1.5 GB	5.5 GB	8 GB card	−1.9% ppl
Q3_K_M	3.9 GB	0.09 GB	0.38 GB	1.5 GB	4.5 GB	8 GB card	−5.4% ppl
Q2_K	3.3 GB	0.09 GB	0.38 GB	1.5 GB	4.0 GB	8 GB card	−15% ppl

Totals add 0.6 GB runtime overhead and an f16 KV cache; "needs" is the smallest discrete card class whose usable memory (after display and driver reserve) leaves 15% headroom. A q8_0 cache roughly halves the KV columns.

Every device, checked

System RAM for offload:

16 32 64 128 256

NVIDIA · GeForce

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
GeForce RTX 5090	32 GB	Q8_0	Q8_0	125K	495	yes	Check
GeForce RTX 5080	16 GB	Q8_0	Q8_0	125K	265	yes	Check
GeForce RTX 5070 Ti	16 GB	Q8_0	Q8_0	125K	247	yes	Check
GeForce RTX 5070	12 GB	Q6_K	Q8_0	125K	186	yes	Check
GeForce RTX 5060 Ti 16 GB	16 GB	Q8_0	Q8_0	125K	124	yes	Check
GeForce RTX 5060	8 GB	Q4_K_M	Q5_K_M	125K	124	tight	Check
GeForce RTX 4090	24 GB	Q8_0	Q8_0	125K	278	yes	Check
GeForce RTX 4080 Super	16 GB	Q8_0	Q8_0	125K	203	yes	Check
GeForce RTX 4080	16 GB	Q8_0	Q8_0	125K	198	yes	Check
GeForce RTX 4070 Ti Super	16 GB	Q8_0	Q8_0	125K	186	yes	Check
GeForce RTX 4070 Ti	12 GB	Q6_K	Q8_0	125K	139	yes	Check
GeForce RTX 4070 Super	12 GB	Q6_K	Q8_0	125K	139	yes	Check
GeForce RTX 4070	12 GB	Q6_K	Q8_0	125K	139	yes	Check
GeForce RTX 4060 Ti 16 GB	16 GB	Q8_0	Q8_0	125K	80	yes	Check
GeForce RTX 4060 Ti	8 GB	Q4_K_M	Q5_K_M	125K	80	tight	Check
GeForce RTX 4060	8 GB	Q4_K_M	Q5_K_M	125K	75	tight	Check
GeForce RTX 3090 Ti	24 GB	Q8_0	Q8_0	125K	278	yes	Check
GeForce RTX 3090	24 GB	Q8_0	Q8_0	125K	258	yes	Check
GeForce RTX 3080 Ti	12 GB	Q6_K	Q8_0	125K	252	yes	Check
GeForce RTX 3080 12 GB	12 GB	Q6_K	Q8_0	125K	252	yes	Check
GeForce RTX 3080	10 GB	Q6_K	Q6_K	125K	210	yes	Check
GeForce RTX 3070 Ti	8 GB	Q4_K_M	Q5_K_M	125K	168	tight	Check
GeForce RTX 3070	8 GB	Q4_K_M	Q5_K_M	125K	124	tight	Check
GeForce RTX 3060 Ti	8 GB	Q4_K_M	Q5_K_M	125K	124	tight	Check
GeForce RTX 3060 12 GB	12 GB	Q6_K	Q8_0	125K	99	yes	Check
GeForce RTX 2080 Ti	11 GB	Q6_K	Q8_0	125K	170	yes	Check
GeForce RTX 2060 12 GB	12 GB	Q6_K	Q8_0	125K	93	yes	Check
GeForce GTX 1080 Ti	11 GB	Q6_K	Q8_0	125K	134	yes	Check

NVIDIA · GeForce laptop

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
GeForce RTX 5090 Laptop	24 GB	Q8_0	Q8_0	125K	247	yes	Check
GeForce RTX 4090 Laptop	16 GB	Q8_0	Q8_0	125K	159	yes	Check
GeForce RTX 4080 Laptop	12 GB	Q6_K	Q8_0	125K	119	yes	Check
GeForce RTX 4070 Laptop	8 GB	Q4_K_M	Q5_K_M	125K	71	tight	Check

NVIDIA · Workstation

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
RTX 6000 Ada	48 GB	Q8_0	Q8_0	125K	265	yes	Check
RTX 5000 Ada	32 GB	Q8_0	Q8_0	125K	159	yes	Check
RTX 4000 Ada	20 GB	Q8_0	Q8_0	125K	99	yes	Check
RTX 2000 Ada	16 GB	Q8_0	Q8_0	125K	62	yes	Check
RTX A6000	48 GB	Q8_0	Q8_0	125K	212	yes	Check
RTX A5000	24 GB	Q8_0	Q8_0	125K	212	yes	Check
RTX A4000	16 GB	Q8_0	Q8_0	125K	124	yes	Check

NVIDIA · Data centre

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
L40S	48 GB	Q8_0	Q8_0	125K	239	yes	Check
L4	24 GB	Q8_0	Q8_0	125K	83	yes	Check
A10	24 GB	Q8_0	Q8_0	125K	166	yes	Check
A100 40 GB	40 GB	Q8_0	Q8_0	125K	429	yes	Check
A100 80 GB	80 GB	Q8_0	Q8_0	125K	563	yes	Check
H100 PCIe	80 GB	Q8_0	Q8_0	125K	552	yes	Check
H100 SXM	80 GB	Q8_0	Q8_0	125K	925	yes	Check
H200 SXM	141 GB	Q8_0	Q8_0	125K	1325	yes	Check
Tesla P40	24 GB	Q8_0	Q8_0	125K	96	yes	Check

AMD · Radeon

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Radeon RX 9070 XT	16 GB	Q8_0	Q8_0	125K	178	yes	Check
Radeon RX 9070	16 GB	Q8_0	Q8_0	125K	178	yes	Check
Radeon RX 7900 XTX	24 GB	Q8_0	Q8_0	125K	265	yes	Check
Radeon RX 7900 XT	20 GB	Q8_0	Q8_0	125K	221	yes	Check
Radeon RX 7900 GRE	16 GB	Q8_0	Q8_0	125K	159	yes	Check
Radeon RX 7800 XT	16 GB	Q8_0	Q8_0	125K	172	yes	Check
Radeon RX 7700 XT	12 GB	Q6_K	Q8_0	125K	119	yes	Check
Radeon RX 7600 XT	16 GB	Q8_0	Q8_0	125K	80	yes	Check
Radeon RX 6900 XT	16 GB	Q8_0	Q8_0	125K	141	yes	Check
Radeon RX 6800 XT	16 GB	Q8_0	Q8_0	125K	141	yes	Check
Radeon RX 6700 XT	12 GB	Q6_K	Q8_0	125K	106	yes	Check

AMD · Radeon Pro

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Radeon Pro W7900	48 GB	Q8_0	Q8_0	125K	239	yes	Check
Radeon Pro W7800	32 GB	Q8_0	Q8_0	125K	159	yes	Check

AMD · Instinct

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Instinct MI210	64 GB	Q8_0	Q8_0	125K	452	yes	Check
Instinct MI300X	192 GB	Q8_0	Q8_0	125K	1463	yes	Check

AMD · Ryzen AI Max

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Ryzen AI Max+ 395 · 128 GB	128 GB	Q8_0	Q8_0	125K	79	yes	Check
Ryzen AI Max+ 395 · 64 GB	64 GB	Q8_0	Q8_0	125K	79	yes	Check

Apple · Apple Silicon

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
M1 · 16 GB	16 GB	Q8_0	Q8_0	125K	21	yes	Check
M1 Pro · 32 GB	32 GB	Q8_0	Q8_0	125K	62	yes	Check
M1 Max · 64 GB	64 GB	Q8_0	Q8_0	125K	124	yes	Check
M1 Ultra · 128 GB	128 GB	Q8_0	Q8_0	125K	247	yes	Check
M2 · 24 GB	24 GB	Q8_0	Q8_0	125K	31	yes	Check
M2 Pro · 32 GB	32 GB	Q8_0	Q8_0	125K	62	yes	Check
M2 Max · 96 GB	96 GB	Q8_0	Q8_0	125K	124	yes	Check
M2 Ultra · 192 GB	192 GB	Q8_0	Q8_0	125K	247	yes	Check
M3 · 24 GB	24 GB	Q8_0	Q8_0	125K	31	yes	Check
M3 Pro · 36 GB	36 GB	Q8_0	Q8_0	125K	46	yes	Check
M3 Max · 64 GB	64 GB	Q8_0	Q8_0	125K	124	yes	Check
M3 Max · 128 GB	128 GB	Q8_0	Q8_0	125K	124	yes	Check
M3 Ultra · 256 GB	256 GB	Q8_0	Q8_0	125K	247	yes	Check
M3 Ultra · 512 GB	512 GB	Q8_0	Q8_0	125K	247	yes	Check
M4 · 32 GB	32 GB	Q8_0	Q8_0	125K	37	yes	Check
M4 Pro · 48 GB	48 GB	Q8_0	Q8_0	125K	84	yes	Check
M4 Pro · 64 GB	64 GB	Q8_0	Q8_0	125K	84	yes	Check
M4 Max · 64 GB	64 GB	Q8_0	Q8_0	125K	127	yes	Check
M4 Max · 128 GB	128 GB	Q8_0	Q8_0	125K	169	yes	Check

Intel · Arc

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Arc B580	12 GB	Q6_K	Q8_0	125K	126	yes	Check
Arc B570	10 GB	Q6_K	Q6_K	125K	105	yes	Check
Arc A770 16 GB	16 GB	Q8_0	Q8_0	125K	155	yes	Check
Arc A750	8 GB	Q4_K_M	Q5_K_M	125K	141	tight	Check

CPU · CPU + system RAM

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
CPU only · DDR4 dual-channel	32 GB	Q8_0	Q8_0	125K	11	yes	Check
CPU only · DDR5 dual-channel	64 GB	Q8_0	Q8_0	125K	20	yes	Check
CPU only · DDR5 8-channel server	256 GB	Q8_0	Q8_0	125K	68	yes	Check