LDLMIn construction

All models, with previews →

Meshchaninov et al. · May 2026Diffusion language model

Self-attention

Pinned implementation and paper LDLM @ e18c42a, not yet public · arXiv:2605.07933v1 · OpenWebText · 1024 · checkpoint, tokenizer and saved statistics not verified · How to read

TrainingGeneration

Sampler › Denoiser › 12 denoiser blocks › Self-attention

Output projection

×

learned

Map the merged features to the attention update.

Δattn=oWO{\htmlData{light=update output-projection}{\Delta_{\mathrm{attn}}}}={\htmlData{light=merge}{o}}W_{O}
in
oo [B batch, T positions, 768 d]
out
Δattn\Delta_{\mathrm{attn}} [B batch, T positions, 768 d]
parameters
589,824 · × 12 blocks = 7,077,888
Parameters

589,824 parameters per instance.

Exact per-instance accounting
ComponentCalculationParameters
Weight matrix768×768768 \times 768589,824
Per-instance total589,824

This projection has no bias parameter.

The reference module holds the weight output by input, [out, in].

Across 12 independent blocks: 589,824×12=7,077,888589{,}824 \times 12 = 7{,}077{,}888 parameters.

Derived from the configured matrix dimensions; checkpoint tensors have not been independently counted.

Sources2

Direct evidence

A block owns two gain-only LayerNorms, a fused query–key–value projection and an output projection without bias, an MLP of the configured kind, and, with modulation, an affine layer from the time condition to six vectors, initialized at zero.

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