Welcome to Torch-Secorder’s documentation!

Torch-Secorder is a PyTorch library for efficient second-order optimization and curvature computation.

Contents:

• Installation
  • Requirements
  • Version Compatibility
  • Installing from PyPI
  • Installing from Source
  • Verifying Installation
• Quickstart Guide
  • Basic Usage
  • Advanced Usage
  • Best Practices
• Core
  • approximate_hvp()
  • batch_jvp()
  • batch_vjp()
  • exact_hvp()
  • flatten_params()
  • full_jacobian()
  • gauss_newton_product()
  • get_layer_curvature_stats()
  • get_param_shapes()
  • get_params_by_module_type()
  • get_params_by_name_pattern()
  • hessian_diagonal()
  • hessian_trace()
  • jvp()
  • model_hessian_diagonal()
  • model_hessian_trace()
  • model_hvp()
  • model_jvp()
  • model_vjp()
  • per_layer_fisher_diagonal()
  • per_layer_hessian_diagonal()
  • unflatten_params()
  • vjp()
  • Hessian-Vector Product (HVP) Module
  • Jacobian-Vector Product (GVP) Module
  • Hessian Diagonal Module
  • Hessian Trace Module
  • Utility Functions
  • Per-Layer Curvature Module
• Approximations
  • empirical_fisher_diagonal()
  • empirical_fisher_trace()
  • gauss_newton_matrix_approximation()
  • generalized_fisher_diagonal()
  • generalized_fisher_trace()
  • Fisher Diagonal Module
  • Fisher Trace Module
  • Gauss-Newton Module
• Analysis Module
  • Eigensolvers
  • Loss Landscape Visualization
• Practical Applications & Tutorials
  • Analysis Examples
  • Optimization Examples
  • General Examples
• Security contact information
• Contributing
  • Development Setup
  • Code Style
  • Documentation
  • Pull Request Process
  • Bug Reports
  • Code of Conduct

Indices and tables

• Index

• Module Index

• Search Page