Wang et al.: DeepNorm residual scaling and initialization

What changed

Combined modified residual normalization with depth-dependent initialization to stabilize very deep Transformer training, including reported thousand-layer experiments.

Scientific scope

Combined modified residual normalization with depth-dependent initialization to stabilize very deep Transformer training, including reported thousand-layer experiments. Source-reported analysis and training results, not an independently audited general convergence theorem. Inherits residual networks and normalization; depth scaling does not alone guarantee useful task performance.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Source-reported analysis and training results, not an independently audited general convergence theorem. Inherits residual networks and normalization; depth scaling does not alone guarantee useful task performance.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• DeepNet: Scaling Transformers to 1,000 Layers
• Wang et al.: DeepNorm residual scaling and initialization — official project source

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction