LFM2.5-8B-A1B

LFM family · 8.0B params

2recorded runs

277.8best tok/s

2verified

Evidence: 2 sourced runs · 1 independent source · first recorded May 30, 2026 · last recorded Aug 20, 2026. What these labels mean.

What the LFM2.5-8B-A1B numbers mean

The fastest verified result on our board is 277.8 output tok/s on an RTX 3080 at Q6_K with a 128000-token context. That is about 3.6 ms per generated token — 500 tokens takes roughly 1.8 seconds. Prompt processing on the same record runs at 9840.7 tok/s, so a 2,000-token prompt is read in about 0.2 seconds. Peak VRAM was measured at 8.1 GB. Across the 2 records we hold for this model the range is 241.1–277.8 tok/s on 2 GPUs, because quantisation, context length and the engine build all move the number.

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

LFM2.5-8B-A1B has 8.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	6.0 GB	8 GB
Q8_0 (8-bit)	10.6 GB	16 GB
FP16 (16-bit)	19.2 GB	24 GB

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Fastest GPUs for this model (best tok/s)

RTX 3080277.8
Radeon AI Pro R9700241.1

Models with a similar footprint

Model	Params	Q4_K_M weights
AceReason-Nemotron-1.1-7B	8B	4.5 GB
DeepSeek-R1-0528-Qwen3-8B	8B	4.5 GB
DeepSeek-R1-Distill-Qwen-7B	8B	4.5 GB
gemma-4-E4B-it	8B	4.5 GB
granite-4.2-8b	8B	4.5 GB
InternVL3-8B	8B	4.5 GB

Ranked by distance from LFM2.5-8B-A1B's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
RTX 3080 · 10 GB	LFM2.5-8B-A1B	Q6_K	llama.cpp b10472	128000	277.8 ✓	9840.7	8.1	src
Radeon AI Pro R9700 · 32 GB	LFM2.5-8B-A1B	MQ4	hipfire 0.2.0+d38a269d	4096	241.1 ✓	242.1	5.4	src