Expand description
2026-09-25: Load GLM-5.3’s MTP block, model.language_model.layers.{num_hidden_layers},
as the body of the draft proposer.
Owner: model-arch weight loader (GLM-5.3). Invariants:
- The block is one DSA mixer and one routed MoE with
mhc: None, the plain-residual path ofcrate::glm5next_layer::Glm5NextLayer. - Its DSA layer uses
attn_layer_idx = 0of the drafter’s own one-layer KV pool (glm5next_mtp_head), never the target’s. Glm5NextWeightLoader::prune_after_loadkeeps this layer’s tensors:is_reuploadedmatches only layer indices belownum_hidden_layers.
Structs§
- Glm5
Next MtpModule - 2026-09-25: The MTP block plus the four tensors around it.
Functions§
- load_
glm5next_ mtp_ module - 2026-09-25: Build the MTP block, or
Nonewhen the checkpoint has nolayers.{num_hidden_layers}.