大模型Technical Report2025

Gemini 2.0 Flash: Efficient Multimodal AI

Gemini Team, Google DeepMindGoogle DeepMind

摘要

We present Gemini 2.0 Flash, an efficient multimodal AI model optimized for speed and quality. Gemini 2.0 Flash achieves strong performance across text, image, and code tasks while significantly reducing inference latency, making it suitable for real-time applications at scale.

multimodal AIefficiencylow latencytext understandingcode generation

技术细节

数据集

Gemini Pretraining Corpus (multimodal)

模型骨架

Gemini 2.0 Flash (efficient multimodal)

编码器

Multi-Modal Encoder (text/image/audio)

解码器

Efficient Multi-Modal Decoder