Module load_layers

Module load_layers 

Source
Expand description

2026-09-25: Builds the DeepSeek-V4.1 layers and the V41Runtime they share: the expert arena and cache, the attention, MoE and engram engines, and the engram hash tables.

Owner: model-arch weight loader. Invariants:

  • Every returned layer holds a clone of the same Arc<V41Runtime>.
  • Loading fails unless hc_mult is in 1..=4, the expert stacks hold num_experts experts and compress_ratios has an entry for every layer.

Functionsยง

load_layers