Meaning
An attention design that compresses key and value vectors into a low-dimensional latent space to slash GPU memory bandwidth demands.
Articles Featuring This Term
Related Terms
- Sensitivity and Specificity Detection vs False-Alarm Rates
- Autonomous Dots Persistent Cloud AI Agents
- Convolutional Neural Network (CNN) Visual Pattern Deep Learning
- Side-Channel Attack Hardware Telemetry Leakage Attack
- RingAttention Distributed Sequence Parallelism
- Mel-Spectrogram
- Autoregressive Model Sequential Next-Token Generator
- Mixture-of-Experts Sparse Routing Neural Network
- Process Reward Model Step-by-Step Logic Evaluator