thushv89/attention_keras
A Keras/TensorFlow implementation of Bahdanau attention for RNN-based sequence models used in neural machine translation.

Velocity · 7d
+0.2
★ / day
Trend
→steady
star history
This repository provides a custom Keras layer implementing the Bahdanau attention mechanism, a key component in sequence-to-sequence models. The implementation integrates with TensorFlow 2.x and supports both unidirectional and bidirectional RNN encoders. It includes example code for neural machine translation tasks, complete with training scripts and utilities for visualizing attention weight heatmaps.