Tag: model efficiency
Browse our exclusive articles!
Self-Routing: Parameter-Free Expert Routing for MoE Models
Discover Self-Routing, a parameter-free expert routing method that boosts MoE model efficiency by using hidden states without a learned router.
Controlling Dynamic Head Pruning in Vision Transformers
Enhance Vision Transformer efficiency with sparse autoencoder latents for dynamic head pruning, boosting accuracy and interpretability in computer vision m...
Diminishing Returns of Early-Exit Decoding in LLMs
Explore why early-exit decoding shows reduced effectiveness in modern large language models and what this means for future NLP research.