LFM2.5-Encoders: Fast Long-Context CPU Inference
Liquid AI released LFM2.5-Encoders, bidirectional encoder models built from LFM2.5 decoders. They support 8,192-token context, achieve 3.7× speedup over ModernBERT on CPU at long context, and excel on GLUE, SuperGLUE, and multilingual tasks.
Liquid AI
Hugging Face blog28.07 · 19:02