Liquid AI's QAD technique lets Q4 models retain 97% of BF16 performance
alexcovo_eth · x · 2026-08-20
Liquid AI released Quantization-Aware Distillation (QAD), a method where the model trains while experiencing 4-bit quantization errors, guided by a high-precision teacher model. This allows the quantized model to compensate for precision loss. Applied to LFM2.5 models (230M-2.6B), the resulting Q40 GGUFs retain roughly 97% of their BF16 performance while maintaining a small footprint.
Related event: Liquid AI's QAD Delivers 4-bit Models Retaining 97% Performance(4 posts)→
More from Research
- Dev critiques compiling fast programs into slow neural nets: fun, but not useful — ctjlewis · 2026-10-16
- New paper uses revealed-preference methods to unpack why AI agents cooperate — brwilder · 2026-10-16
- Logit lens on a Beat Saber finetune reveals it reuses language grammar circuits — ctjlewis · 2026-10-16
- 128×128 record stands: how intermediate-work experiments foreshadowed reasoning models — ctjlewis · 2026-10-16
- TeamFusion: preference-based AI proxies debate to preserve minority views in team decisions — burkov · 2026-10-16
- MIT's C-GMT gives humanoid robots whole-body compliance without sacrificing motion tracking — MIT_CSAIL · 2026-10-16