>
G
Model Details
Introduction
Benchmarking
Notes: Training compute imputed to be 1.58e25 FLOP from benchmark scores. https://colab.research.google.com/drive/1r3pUMhB7Kh0Gls9eG-v_XefWrye9fVQR?usp=sharing
Training
Training Code AccessibilityAPI access: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models
HardwareGoogle TPU v4
Parameters
Notes: MoE architecture
Authors
Gemini Team
Vision
Video
Speech