📅 14/02/2026 à 14:17

Moonshine, an encoder-decoder transformer architecture for speech recognition, uses Rotary Position Embedding, reducing compute requirements without decreasing accuracy.

📄 Voir l'article original →