LDLMIn construction

All models, with previews →

Meshchaninov et al. · May 2026Diffusion language model

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

Training Generation

GPT-2 Smallhidden state −3124M weightsstandardized Open →

Latent encoder Open →

z₀ Decoder noise

+\,3\,\epsilon

Latent decoder Open →

Token decoder Open →

L Token cross-entropy

\mathcal L_{w}=\operatorname{CE}(L,\,I)

Feature reconstruction

\mathcal L_{h}=\operatorname{MSE}(\hat h,\,h)

z_t=\alpha_t z_0+\sigma_t\epsilon

z_t t z₀

Denoiser Open →

z₀ Diffusion loss

\mathcal L_{\mathrm{diff}}=\operatorname{MSE}(\hat z_0,\,z_0)

learns from

\mathcal L_{h}\cdot\mathcal L_{\mathrm{diff}}

learns from

\mathcal L_{\mathrm{diff}}

learns from

\mathcal L_{h}

learns from

\mathcal L_{w}

joint objective: the three losses are added, and one backward pass trains the four networksfrozenGPT-2 Small →∇ × γ ∇ × 0

Detached features

tensor

The reconstructed features, with the gradient stopped.

shape: $\hat h$ [B batch, T positions, 768 d]

Sources

Contextual evidence from Stop-gradient

Reconstruction: Gaussian noise of fixed standard deviation is added to the latents, the latent decoder runs, the token decoder reads its output detached, and the cross-entropy against the token IDs is taken; the latent decoder’s output is also compared with the standardized features by a mean squared error.

The implementation, pinned at revision e18c42acdfb7, not yet public

Contextual evidence from Stop-gradient

Joint training: decoder noise of standard deviation 3; weights 1 for the three losses; the diffusion gradient into the encoder rises from 0.001 over 50,000 steps by a straight-through scaling; the token decoder’s input is detached; AdamW at 3·10⁻⁴ for 1,000,000 steps of 512 sequences, gradients clipped at norm 1.

The implementation, pinned at revision e18c42acdfb7, not yet public

Contextual evidence from Token decoder

The token decoder: a rotary table, the configured number of blocks, and a linear readout to the vocabulary without bias.

The implementation, pinned at revision e18c42acdfb7, not yet public

Contextual evidence from Token decoder

The token decoder’s blocks: no softcap, no time conditioning, the gated MLP with the configured multiplier.

The implementation, pinned at revision e18c42acdfb7, not yet public

Contextual evidence from Token decoder

The latent decoder: 12 heads of 64 features, query and key normalization, softcap 50, 6 layers, MLP multiplier 4; the token decoder has 3 blocks.

The implementation, pinned at revision e18c42acdfb7, not yet public

Contextual evidence from Token decoder

The feature model is GPT-2, frozen; the hidden state at index −3 is extracted; the encoder and the decoder are of the Perceiver kind; the file’s vocabulary size and mask token ID are defaults that the trainer overwrites with the tokenizer’s; there are as many latents as positions, of 768 features.

The implementation, pinned at revision e18c42acdfb7, not yet public