REPLUG: Retrieval-Augmented Black-Box Language Models

Benchmark Model Rank Results
question-answering-on-natural-questionscode-davinci-002 175B + REPLUG LSR (few-shot)#10EM: 45.5
question-answering-on-natural-questionscode-davinci-002 175B + REPLUG (few-shot)#12EM: 44.7
question-answering-on-triviaqacode-davinci-002 175B + REPLUG LSR (Few-Shot)#8EM: 77.3
question-answering-on-triviaqacode-davinci-002 175B + REPLUG (Few-Shot)#10EM: 76.8