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

MLP

Training Generation

x_mlp[B, T, 768]Expansion

h=x_{\mathrm{mlp}}W_{1}+b_{1}

h[B, T, 3072]GELU

g=\operatorname{GELU}(h)

g[B, T, 3072]Contraction

\Delta_{\mathrm{mlp}}=gW_{2}+b_{2}

Δ_mlp[B, T, 768]

Denoiser › 12 denoiser blocks › MLP

MLP update

tensor

The update handed to the time gate.

shape: $\Delta_{\mathrm{mlp}}$ [B batch, T positions, 768 d]

Sources

Contextual evidence from Gate

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

Contextual evidence from Gate

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

Contextual evidence from Contraction

The standard MLP: an affine layer, GELU with the tanh approximation, an affine layer.

The implementation, pinned at revision e18c42acdfb7, not yet public