Liquid AI releases QAD Q4_0 checkpoints for four LFM2.5 models
Liquid AI has released updated Q4_0 GGUF checkpoints for LFM2.5-230M, LFM2.5-350M, LFM2.5-1.2B-Instruct, and LFM2.5-2.6B. The checkpoints were produced with Quantization-Aware Distillation, which transfers capabilities from a high-precision teacher to a quantized student model. They are designed to retain Q4_0 memory use and throughput while reducing the typical quality loss from quantization.
In benchmark testing covering reasoning, instruction following, tool use, and agentic capabilities, the four checkpoints retained 97.1%, 96.5%, 97.4%, and 96.6% of their corresponding BF16 baseline performance. The evaluation included GPQA Diamond, MMLU-Pro, IFEval, IFBench, Multi-IF, BFCLv4, and a model-scale-appropriate math test. Results were averaged over five repetitions.
On MacBook Pro, NucBox EVO-X2, Samsung Galaxy S26 Ultra, and Raspberry Pi 5, the smaller QAD checkpoints matched Q5_K_M quality within evaluation variance while delivering 4–33% higher decode throughput. The 1.2B and 2.6B versions matched Q4_K_M quality with 3–14% higher throughput. The files are available on Hugging Face for use with llama.cpp or other GGUF-compatible runtimes.
Written by Sakhanda Wire AI from reporting by Hugging Face. Facts, quotes and full context are in the original article below.