BWhat changed
Co-designed gated short-convolution/GQA backbones through hardware-in-the-loop search and combined them with tempered decoupled top-k distillation, releasing dense, sparse and multimodal edge-oriented variants.
Scientific scope
Co-designed gated short-convolution/GQA backbones through hardware-in-the-loop search and combined them with tempered decoupled top-k distillation, releasing dense, sparse and multimodal edge-oriented variants. Convolution, GQA, NAS, distillation and ColBERT are inherited; integrated constrained design/training/deployment is the delta. Liquid branding is not evidence of a continuous-time liquid-neural-network architecture. CPU-specific latency gains do not generalize to all devices. Front requests LiquidAI citation; contributor appendix not promoted.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Convolution, GQA, NAS, distillation and ColBERT are inherited; integrated constrained design/training/deployment is the delta. Liquid branding is not evidence of a continuous-time liquid-neural-network architecture. CPU-specific latency gains do not generalize to all devices. Front requests LiquidAI citation; contributor appendix not promoted.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- LFM2 Technical Report
- Liquid AI: hardware-constrained hybrid model and distillation integration — author/institutional project
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction