Multi-Token Prediction Done Right
Open paper
Benchmark
Model
Rank
Results
language-modelling-on-lambada
Soft-MTP (1.2B, zero-shot)
#54
Accuracy: 46.48
question-answering-on-piqa
Soft-MTP (1.2B, zero-shot)
#142
Accuracy: 72.03
Rank counts only results with a code link.