#LiquidAI

LFM2.5-24B-A2B Hands-On: Holding Up Under Heavy Load in Pi Agent

Reddit user tests LiquidAI's new LFM2.5-24B-A2B: ~95 tps on AMD W7900 via llama.cpp+ROCm, Q8 quantization at 30GB VRAM with 128k context, stable in Pi Agent mechanical tasks, a reliable small model for dual-GPU collaboration.