LDLMIn construction

All models, with their blocks →

Size
306M trained
Published
Meshchaninov et al. · May 2026
Line
Diffusion language model
View
Architecture
Source

LDLM @ e18c42a, not yet public · arXiv:2605.07933v1 · OpenWebText · 1024 · checkpoint, tokenizer and saved statistics not verifiedHow to read

Token-decoder block

TrainingGeneration

Token decoder › 3 blocks

Add MLP update

×

no parameters

Add the mechanism’s update to the residual stream element by element.

yn+1=yn′+Δmlp{\htmlData{light=block-output mlp-add}{y_{n+1}}}={\htmlData{light=residual attention-add}{y'_n}}+{\htmlData{light=mlp}{\Delta_{\mathrm{mlp}}}}
in
yn′y'_n [B batch, T positions, 768 d]Δmlp\Delta_{\mathrm{mlp}} [B batch, T positions, 768 d]
out
yn+1y_{n+1} [B batch, T positions, 768 d]
Sources2

Direct evidence

The residual update is scale · dropout(x) added to the residual; modulation is x · (1 + scale) + shift.

The implementation, pinned at revision e18c42acdfb7, not yet public

Direct evidence

A block normalizes, modulates by the time condition’s shift and scale, projects to queries, keys and values, rotates queries and keys, attends over all positions with softcap, and adds the projected output scaled by a gate; the MLP branch does the same with its own shift, scale and gate.

The implementation, pinned at revision e18c42acdfb7, not yet public