LFM2.5-350M

LFM family · 0.3B params

5recorded runs

466.0best tok/s

5verified

Evidence: 5 sourced runs · 1 independent source · first recorded Aug 10, 2026 · last recorded Aug 10, 2026. What these labels mean.

What the LFM2.5-350M numbers mean

The fastest verified result on our board is 466.0 output tok/s on an Radeon AI Pro R9700 at MQ4 with a 4096-token context. That is about 2.1 ms per generated token — 500 tokens takes roughly 1.1 seconds. Across the 5 records we hold for this model the range is 207.7–466.0 tok/s on 5 GPUs, because quantisation, context length and the engine build all move the number.

How much VRAM does LFM2.5-350M need?

LFM2.5-350M has 0.3B 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)	1.0 GB	8 GB
Q8_0 (8-bit)	1.2 GB	8 GB
FP16 (16-bit)	1.6 GB	8 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)

Models with a similar footprint

Model	Params	Q4_K_M weights
LFM2-350M	350M	0.2 GB
MobileLLM-350M	350M	0.2 GB
SmolLM2-360M-Instruct	360M	0.2 GB
LFM2.5-230M	230M	0.1 GB
Qwen2.5-0.5B-Instruct	500M	0.3 GB
SmolLM2-135M-Instruct	135M	0.1 GB

Ranked by distance from LFM2.5-350M's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
Radeon AI Pro R9700 · 32 GB	LFM2.5-350M	MQ4	hipfire 0.3.0+439233d92959	4096	466.0 ✓	—	—	src
RX 9070 XT · 16 GB	LFM2.5-350M	MQ4	hipfire 0.3.0+e734d6c8561d	4096	437.7 ✓	—	—	src
RX 7900 XTX · 24 GB	LFM2.5-350M	MQ4	hipfire 0.3.0+b0bcc3f91506	4096	398.4 ✓	—	—	src
RX 6900 XT · 16 GB	LFM2.5-350M	MQ4	hipfire 0.3.0+b0bcc3f91506	4096	275.8 ✓	—	—	src
RX 5700 XT · 8 GB	LFM2.5-350M	MQ4	hipfire 0.3.0+b0bcc3f91506	4096	207.7 ✓	—	—	src