Pointer Head
A neural network output layer structure that does not predict a distribution over a fixed vocabulary, but instead assigns probability weights to candidate positions or candidate sets already present in the input, allowing the model to obtain the full probability distribution of all answers in a single forward pass without token-by-token decoding.
Timeline (last 90 days)
The pointer head is a lightweight classification head that maps the hidden state directly onto a distribution over predefined answer options, making the output fall within a closed label set rather than letting the model freely generate answer text.
Drex DLM's pointer head directly scores the answer options rather than decoding text, producing a calibrated probability distribution over the fixed options in a single forward pass.
All Facts (2)
The pointer head is a lightweight classification head that maps the hidden state directly onto a distribution over predefined answer options, making the output fall within a closed label set rather than letting the model freely generate answer text.
50%UnverifiedDrex DLM's pointer head directly scores the answer options rather than decoding text, producing a calibrated probability distribution over the fixed options in a single forward pass.
50%