Learning to Read Out: Unembedding Dynamics in Language Model Pretraining
Pythia encodes task information in hidden states before its unembedding readout can express it.
1 stories, newest first · one card per story
Pythia encodes task information in hidden states before its unembedding readout can express it.
Today 2.2K readers · 2.7K views · 401 reads · all time 108K views