Primitives

Core abstractions used throughout the framework.

Zero-copy flat-tensor view

Krum’s primitives are built around a zero-copy flat-tensor view of PyTorch modules. Instead of working with nested parameter structures, aggregators and attacks operate on a single 1-D vector representation of the model state.

This design provides:

• Efficiency: No data copying on every access — the flat tensor shares memory with the module’s parameters and gradients

• Simplicity: Aggregators work with a single (d,) tensor instead of iterating over nested structures

• Flexibility: Gradients can be read, modified, and written back in place

The Model wrapper encapsulates this behavior, exposing .parameters and .gradients as flat tensors that share the underlying buffer.

Core Abstractions:

• Model