transformer-original-implementation-tensorflow

IN premise — summaries/2026-08-24/wiki-Transformer_deep_learning_architecture-chunk-1.md

Created 2026-08-24T17:11:24+00:00

The original Transformer reference implementation (Vaswani et al., 2017) was written in TensorFlow.

Summary

The reference code that the original 2017 "Attention Is All You Need" authors released to let others reproduce their architecture was built in TensorFlow, not PyTorch. This matters because it anchors the provenance of the Transformer: any claim about how the system was first implemented, debugged, or reproduced traces back to that TensorFlow codebase as the ground truth.