microsoft / deep-language-networks

We view Large Language Models as stochastic language layers in a network, where the learnable parameters are the natural language prompts at each layer. We stack two such layers, feeding the output of one layer to the next. We call the stacked architecture a Deep Language Network - DLN
MIT License
91 stars 13 forks source link

added script for one-layer experiments #7

Closed matheper closed 1 year ago