Jan 2, 2026 · Artificial Intelligence

How Manifold-Constrained Hyper-Connections Boost LLM Performance with Minimal Overhead

DeepSeek's new mHC architecture projects residual connections onto a manifold, enabling a 6.7% training cost increase for 27B models while delivering significant stability and downstream performance gains over traditional residual and hyper‑connection designs.

Deep LearningLLMManifold Optimization

0 likes · 13 min read

How Manifold-Constrained Hyper-Connections Boost LLM Performance with Minimal Overhead

Alimama Tech

Nov 3, 2021 · Artificial Intelligence

Curvature Learning Framework (CurvLearn): A TensorFlow‑Based Library for Non‑Euclidean Deep Learning

CurvLearn is a TensorFlow-based open-source library enabling deep learning on curved manifolds (hyperbolic, spherical, mixed) with manifold implementations, Riemannian operations, optimizers, and distributed training, and it has been applied to recommendation, graph, and NLP tasks while providing custom ANN tools and practical training tips.

Curvature LearningDeep LearningManifold Optimization

0 likes · 13 min read

Curvature Learning Framework (CurvLearn): A TensorFlow‑Based Library for Non‑Euclidean Deep Learning