RT @ColesThermoAI: Computing matrix exponentials from thermodynamic fluctuations (equation below) ... is now published!

RT @gm8xx8: Distillation Scaling Laws  This study presents a distillation scaling law to estimate student model performance based on com…

The MoE-Empowered Edge LLMs Deployment: Architecture, Challenges, and Opportunities paper: CoEL, an MoE-based framework, optimizes resource allocation through dynamic deployment, compression, and continuous updates. It enables efficient intra- and inter-device collaboration while reducing memory usage. The paper outlines key challenges.

code: paper:

Tulu-3.1-8B (trained with GRPO)

@samsja19 lfg

models: paper: