LFM2 8B A1B

About the LFM (Liquid) family

Liquid AI's non-Transformer foundation models (LFM2/2.5 series, 1.2B–24B including MoE 'A1B/A2B' and vision variants) built for on-device speed and memory efficiency.

How this one is configured

This configuration mixture-of-experts with ~8B total / ~1B active parameters per token.

Who makes it

More from LFM (Liquid)

• LFM 40B
• LFM2 1.2B
• LFM2 2.6B
• LFM2 24B A2B
• LFM2.5-1.2B-Instruct
• LFM2.5-1.2B-Thinking
• LFM2.5-8B-A1B
• LFM2.5-VL-1.6B