AnalysisAI ModelsNovember 3, 2026

LFM3-1.7B quantized to 3.6 GB VRAM lossless, 2.2 GB with degradation

Read original source →reddit.com

Community quantization report puts LiquidAI's LFM3-1.7B at 3.6 GiB VRAM for a quasi-lossless quant versus 7.1 GiB for MiniCPM5-2B. The report cites LiquidAI benchmarks showing LFM3-1.7B scoring higher than MiniCPM5-2B across the board with 2x the context.

1 source

More stories today

Open the live feed