llamaperf

LFM3 2B VL

on M1 8GB · mlx-vlm · 32,768 ctx

Tone: negativeApple hardware
Nov 3, 2026
Throughput
1.0 t/s gen
Quant
OptiQ-4bit (MLX)
System RAM
8 GB

Use cases

vision

Summary

User reports LFM3-VL-2B running at roughly 1 tok/s on a Mac Mini M1 8GB, and says it is worse than LFM2.5-VL-3B. Setup is MLX-VLM with OptiQ 4-bit weights at 32768 context and temperature 0.4. The speed is given only as an approximation (~1 tok/s), and the user notes the model was tested at 16K context.

Compare this run

Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.

LFM3 2B at 4-bit by hardware

Bars are medians on one scale, 0 to 1.0 tokens/s. Press ▶ to watch text arrive at a row's speed.

Models most run on the M1 8GB

Bars are medians on one scale, 0 to 22.0 tokens/s. Press ▶ to watch text arrive at a row's speed.

Everything reported on the M1 8GB →Every LFM3 report →Estimate LFM3 on this card →