LoGAH: Predicting 774-Million-Parameter Transformers using Graph HyperNetworks with 1/100 Parameters.
Xinyu ZhouBoris KnyazevAlexia Jolicoeur-MartineauJie FuPublished in: CoRR (2024)
Keyphrases
- parameter values
- input parameters
- parameter settings
- parameter tuning
- parameter space
- parameter set
- design parameters
- optimal parameters
- control parameters
- real world
- parameters estimation
- single parameter
- graph theory
- parameter adjustment
- graph model
- random walk
- maximum likelihood
- network parameters
- connected components
- high quality
- graph representation
- directed graph
- kernel parameters
- parameter estimates
- initial conditions
- data sets
- artificial neural networks
- tens of thousands
- parameter estimation
- transfer function
- structured data
- sensitivity analysis