RecurrentGemma: Superando los Transformers para Modelos de Lenguaje Abiertos Eficientes

RecurrentGemma: Moving Past Transformers for Efficient Open Language Models

April 11, 2024

Autores: Aleksandar Botev, Soham De, Samuel L Smith, Anushan Fernando, George-Cristian Muraru, Ruba Haroun, Leonard Berrada, Razvan Pascanu, Pier Giuseppe Sessa, Robert Dadashi, Léonard Hussenot, Johan Ferret, Sertan Girgin, Olivier Bachem, Alek Andreev, Kathleen Kenealy, Thomas Mesnard, Cassidy Hardin, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, Pouya Tafti, Armand Joulin, Noah Fiedel, Evan Senter, Yutian Chen, Srivatsan Srinivasan, Guillaume Desjardins, David Budden, Arnaud Doucet, Sharad Vikram, Adam Paszke, Trevor Gale, Sebastian Borgeaud, Charlie Chen, Andy Brock, Antonia Paterson, Jenny Brennan, Meg Risdal, Raj Gundluru, Nesh Devanathan, Paul Mooney, Nilay Chauhan, Phil Culliton, Luiz GUStavo Martins, Elisa Bandy, David Huntsperger, Glenn Cameron, Arthur Zucker, Tris Warkentin, Ludovic Peran, Minh Giang, Zoubin Ghahramani, Clément Farabet, Koray Kavukcuoglu, Demis Hassabis, Raia Hadsell, Yee Whye Teh, Nando de Frietas

cs.AI

Resumen

Presentamos RecurrentGemma, un modelo de lenguaje abierto que utiliza la novedosa arquitectura Griffin de Google. Griffin combina recurrencias lineales con atención local para lograr un rendimiento excepcional en tareas de lenguaje. Posee un estado de tamaño fijo, lo que reduce el uso de memoria y permite una inferencia eficiente en secuencias largas. Ofrecemos un modelo preentrenado con 2B parámetros no incrustados, junto con una variante ajustada por instrucciones. Ambos modelos alcanzan un rendimiento comparable a Gemma-2B a pesar de haber sido entrenados con menos tokens.

English

We introduce RecurrentGemma, an open language model which uses Google's novel Griffin architecture. Griffin combines linear recurrences with local attention to achieve excellent performance on language. It has a fixed-sized state, which reduces memory use and enables efficient inference on long sequences. We provide a pre-trained model with 2B non-embedding parameters, and an instruction tuned variant. Both models achieve comparable performance to Gemma-2B despite being trained on fewer tokens.