LiquidAI has published QAD Q4_0 GGUF checkpoints for four LFM2.5 variants: LFM2.5-230M, LFM2.5-350M, LFM2.5-1.2B-Instruct and LFM2.5-2.6B. The models are trained with Quantization-Aware Distillation (QAD), a process that distills a high-precision teacher into a quantized student to reduce the quality loss typically seen after 4-bit quantization.
Benchmark methodology and results
According to LiquidAI, the team evaluated the QAD Q4_0 checkpoints against post-training quantized (PTQ) GGUFs using a suite that covers reasoning, instruction following, tool use and agentic tasks: GPQA Diamond, MMLU-Pro, IFEval, IFBench, Multi-IF and BFCLv4. Math evaluations were GSM8K for the 230M and 350M models, and AIME25 for the 1.2B-Instruct and 2.6B models. Scores are reported as the mean over five repeats, with a BF16 GGUF serving as the in-format ceiling.
LiquidAI reports that the QAD checkpoints recover most of the accuracy lost to quantization, retaining 97.1%, 96.5%, 97.4% and 96.6% of their respective BF16 baselines for the 230M, 350M, 1.2B-Instruct and 2.6B models.
Performance on edge hardware and compatibility
Throughput tests were run on MacBook Pro and NucBox EVO-X2 (GPU inference) and on Samsung Galaxy S26 Ultra and Raspberry Pi 5 (Arm CPU). BF16 and F16 baselines are included as full-precision references where profiled. For LFM2.5-230M and -350M, LiquidAI states the QAD Q4_0 checkpoints match Q5_K_M quality within evaluation variance while delivering 4–33% higher decode throughput. For LFM2.5-1.2B-Instruct and -2.6B, the QAD Q4_0 checkpoints match Q4_K_M quality at 3–14% higher throughput. The report also notes parity with Unsloth’s UD-Q4_K_XL where applicable (230M and 1.2B).
Files are provided as GGUF Q4_0 artifacts compatible with runtimes such as llama.cpp. Example invocation shown by LiquidAI:
llama-cli -hf LiquidAI/LFM2.5-350M --hf-file LFM2.5-350M-QAD-Q4_0.gguf -p "What is C. elegans?"
The QAD GGUFs are available on Hugging Face for LFM2.5-230M, LFM2.5-350M, LFM2.5-1.2B-Instruct and LFM2.5-2.6B.
Original source: Hugging Face Blog