OpenNN v8.0.0

OpenNN 8.0.0

New Features

Convolutional Neural Networks (CNN)

• Added the ImageClassificationNetwork class.
• Includes support for configurable convolutional blocks, pooling layers, and dense classification layers.

Text Classification Networks

• Added the TextClassificationNetwork class featuring embedding, multi-head attention, and pooling layers.
• Fully supports both binary (Sigmoid) and multi-class (Softmax) classification tasks.

Transformer Architecture

• Added a new encoder-decoder Transformer class designed for sequence-to-sequence tasks.
• Features multi-head self-attention, cross-attention, layer normalization, and feed-forward layers.

CUDA Support

• Full Acceleration: All new architectures (CNNs, Text Classification, and Transformers) include complete CUDA implementations with cuDNN and cuBLAS acceleration.
• Performance Optimization: Features asynchronous batch prefetching combined with multi-threaded data pipelines for significantly improved GPU utilization.