À suivreGuide suivant
Attention fenêtre coulissante
IA linguistique
GUIDE IA du langage
Attention combines information from allowed token positions using learned scores and weights.
Écoutez ce guideNarré par l'IA · ~3 min
It is a core transformer operation, while attention masks and model architecture determine which positions can contribute to each representation.
The 2017 Attention Is All You Need paper introduced the transformer architecture built around attention rather than recurrence. Attention can be described through queries, keys, and values. A query is compared with keys to produce scores, and normalized weights combine the corresponding value vectors. Self-attention uses representations from the same sequence, while cross-attention can obtain keys and values from a separate encoded sequence. The allowed positions are important: a causal decoder mask prevents a position from using future positions during autoregressive training or generation. It is therefore inaccurate to say that every self-attention layer always reads every token without restrictions. Scaled dot-product attention divides query-key dot products by the square root of the key dimension before softmax. This scaling addresses how score magnitudes grow with dimension under the paper's assumptions. Softmax produces normalized weights over allowed positions, and the weighted value sum becomes the attention output. In a simple scalar illustration, values two and six with weights three-quarters and one-quarter combine to three. Multi-head attention performs several learned projections and attention operations before combining their outputs. It provides capacity to represent different relationships or subspaces without guaranteeing that every head has a unique, human-readable semantic role. Dense full attention involves pairwise interactions that scale quadratically with sequence length. Efficient implementations and alternative attention patterns address different costs; their benefits need measurement for a particular workload.
Using decoder cross-attention to combine information from an encoded source sequence.
Applying causal self-attention when predicting text without looking at later positions.
Comparing learned query-key scores before weighting value vectors.
Evaluating attention implementations for memory use and runtime at a chosen sequence length.
Attention research examines implementations, sparse patterns, shared key-value projections, and hybrid architectures. These techniques change different parts of the computation, so comparisons should report which attention rule is used and what costs are measured. Faster execution or lower memory use should be assessed alongside task quality and the relevant sequence lengths. A model can retain attention while changing its layout or implementation; identifying that distinction helps explain what an optimization actually improves.
Newsletter gratuite
Trois histoires d'IA vérifiées chaque matin de la semaine, écrites dans un anglais simple. Gratuit pour toujours, sans publicité.
Un e-mail chaque jour de la semaine. Désabonnez-vous en un clic. Nous ne vendons ni ne partageons jamais votre adresse.
Testez-vous
Commentaires instantanés sur chaque réponse et un certificat partageable avec une pièce d'identité vérifiable une fois que vous avez réussi un cours.
Soutenez l’enseignement gratuit de l’IA. AI Understanding est une organisation à but non lucratif 501(c)(3) : pas de publicité, pas de paywall, jamais. Faire un don
Attention combines information from allowed token positions using learned scores and weights. It is a core transformer operation, while attention masks and model architecture determine which positions can contribute to each representation.
Attention uses weighted values, with the permitted context determined by masks and architecture.
« L'attention est tout ce dont vous avez besoin » (Vaswani et al., 2017) ont proposé le transformateur, qui repose sur l'attention plutôt que sur la récurrence.
Chaque jeton produit une requête, une clé et une valeur ; les requêtes sont comparées à des clés pour pondérer les valeurs.
La mise à l'échelle par la racine carrée de la dimension clé empêche les produits scalaires volumineux qui pousseraient softmax dans de minuscules gradients, maintenant ainsi l'entraînement stable.
Dense attention involves pairwise token interactions. Masks restrict permitted positions, but full or causal dense attention still has quadratic interaction growth; the mask changes permitted pairs rather than removing that order of growth.
Continuez à apprendre
Plus de guides sélectionnés pour ce sujet
À suivreGuide suivant
Attention fenêtre coulissante
IA linguistique