vortex_torch.cache

Cache-side operator API.

This module exposes the core primitives used on the cache path:

• Context: Runtime context carrying layout, paging, and auxiliary metadata.

• Reductions: Mean, Max, Min, L2Norm for per-page / per-request aggregation.

• Matrix–matrix/vector ops: GeMM for generalized page-wise matmul on cached tensors.

• Unary elementwise ops: Relu, Silu, Sigmoid, Abs, Add_Mul.

• Binary elementwise ops: Maximum, Minimum, Multiply, Add.

These building blocks are typically used inside vFlow cache update pipelines (e.g., to maintain centroids, envelopes, or other summaries).

Modules

context
elementwise
elementwise_binary
fill
matmul
reduce