AI BriefingLiquid AI Releases LFM2.5 Q4_0
Key point
Liquid AI has released the LFM2.5 Q4_0 models, which minimize quality degradation through Quantization-Aware Distillation (QAD).
Details
Liquid AI has released Q4_0 GGUF checkpoints for the LFM2.5 series (230M, 350M, 1.2B-Instruct, 2.6B). These models apply Quantization-Aware Distillation (QAD) to transfer knowledge from high-resolution teacher models to quantized student models.
Unlike traditional Post-Training Quantization (PTQ), the QAD checkpoints recover 97% of the average accuracy of the BF16 baseline while maintaining the low memory footprint and high throughput characteristic of Q4_0. Benchmark results show that the 230M and 350M models approach Q5_K_M quality with 4–33% higher decoding throughput.
Performance has been verified on various edge hardware, including MacBook Pro, NucBox, Galaxy S26 Ultra, and Raspberry Pi 5, and the models are ready for use in runtimes that support GGUF Q4_0 artifacts, such as llama.cpp.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Needle 2 - A 14MB Agentic LLM for Smartphones, Wearables, Smart Home, and Robots