fairseq2.nn

The neural network module provides building blocks and utilities for creating transformer-based models and other neural network architectures. It includes the unified batch layout system, core neural network layers, and various utilities for efficient neural network operations.

• Batch Layout
• Embeddings
• Position Encoders
• Normalization Layers
• Projection Layers
• Residual Connections
• Incremental State
• fairseq2.nn.utils

See Also

• fairseq2.models - Pre-trained model architectures

• fairseq2.data - Data processing pipelines