Group-aware SSM Elastification
A method that preserves structural constraints in model compression while optimizing for multiple scales.
Category: Concepts
Heterogeneous MLP Elastification
A method for optimizing multi-layer perceptrons (MLPs) to enhance model efficiency across different configurations.
Mamba-Attention
A hybrid attention mechanism that enhances the efficiency and performance of large language models.
Pruning
A technique used to reduce the size of a neural network by removing weights or neurons that contribute little to the model’s performance.
Nemotron Elastic
A framework for building reasoning-oriented large language models that incorporates multiple nested submodels optimized for different deployment configurations.
Robust Majority Voting
An algorithmic approach that aggregates votes from multiple sources in a way that is resilient to adversarial manipulation or noise.
Graph Bisection Algorithm
An algorithm that partitions the vertices of a graph into two disjoint subsets while minimizing the number of edges between the subsets.
Normalized Discounted Cumulative Gain
A metric that evaluates the effectiveness of a retrieval system based on the graded relevance of the retrieved documents.
Kesten-Stigum Threshold
A critical threshold in community detection problems that separates the regimes where community structure can be reliably detected from where it cannot.
Continuous Control
A type of reinforcement learning task where the agent must make decisions in a continuous action space, often used in robotics and simulation environments.