API reference

The API reference is generated from public Python modules and is kept separate from the conceptual documentation. Use these pages when you need signatures, parameter names, class members, shape contracts, dtype contracts, and error boundaries. Use the concept and backend references when you need architectural context or route-selection details.

The public surface is grouped by semantic feature:

Section	Use it for	Coordinate effect
Core API	Containers, coordinate identity, relation metadata, quantized weights, and shared type helpers.	Defines coordinate ownership and metadata used by every operation.
Operations API	Functional sparse operators such as convolution, pooling, feature transforms, sparse algebra, relation builders, and point/voxel helpers.	Each operation documents whether it preserves, changes, aligns, or generates sparse coordinates.
Neural network modules	mlx.nn.Module wrappers over sparse operations.	Mirrors the corresponding functional operation.
Diagnostics API	Stable backend diagnostics through mlx_lattice.backend_info.	Does not mutate sparse tensors.

• Core API
  • Top-level package
  • Sparse tensor
  • Coordinate management
  • Coordinate utilities
  • Point/voxel core
  • Relations
  • Quantized weights
  • Types
• Operations API
  • Operation map
• Neural network modules
  • Module map
  • Coordinate behavior
• Diagnostics API
  • Related pages