Liquid AI released updated 4-bit GGUF checkpoints trained with Quantization-Aware Distillation for the LFM2.5 model family.

Key Details

• Released QAD Q4_0 checkpoints for LFM2.5-230M, LFM2.5-350M, LFM2.5-1.2B-Instruct, and LFM2.5-2.6B
• Trained using Quantization-Aware Distillation (QAD) to distill high-precision teacher models into quantized student models
• Recovered 97% of BF16 average accuracy lost to quantization
• Maintained native Q4_0 memory footprint and decode speed with 3-33% higher throughput over comparable quality baselines
• Available on Hugging Face and compatible with llama.cpp runtimes

In This Story