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

Key and value heads

Training Generation

The latents’ queries, keys and valuesp^a p^a The features’ keys and valuesp^b p^b Take the latents’ own keys Take the features’ keys Join keysalong positions Split heads Key RMSNorm

\lVert x\rVert_2/\sqrt d

K[B, 12, 2T, 64]Take the latents’ own values Take the features’ values Join valuesalong positions Split heads V[B, 12, 2T, 64]

Latent encoder › 6 encoder blocks › Perceiver attention › Key and value heads

Take the features’ values

no parameters

The second 768 features of the fused projection’s output are the features’ values.

{\htmlData{light=take-v-context}{{\htmlClass{vl-stream-v}{v^{b}}}}}=p^{b}_{769:1536}

in: $p^{b}$ [B batch, T positions, 1536]
out: $v^{b}$ [B batch, T positions, 768 d]

Sources1

Direct evidence

Both streams are normalized; queries come from one stream, keys and values from both, concatenated along positions; queries and keys are normalized per head; attention runs over all positions with softcap; the heads are merged and projected.

The implementation, pinned at revision e18c42acdfb7, not yet public