Module weights 

The GLM MLP weight contract, already sharded for this rank.

Structs

Glm5NextDenseMlpWeights

A dense SwiGLU MLP in BF16 — layers 0..first_k_dense_replace, and the shared expert of every routed layer. Same shape, same kernels, different widths.

Glm5NextExpertPtrTable

Device-side pointer tables for ONE projection across the full expert set.

Glm5NextExpertWeights

One routed expert. NVFP4, owned whole by one EP rank — never split further.

Glm5NextMoePtrTables

The three projections’ pointer tables for one routed site.

Glm5NextMoeWeights

A routed MoE site’s weights for this rank.

Nvfp4Proj

One NVFP4 projection as ModelOpt stores it: packed e2m1 pairs, per-block e4m3 scales, and one global f32 scale.