Cola-DLM
ByteDance-Seed

The codebase of Cola DLM
- Hierarchical latent-variable model. ColaTextVAEModel provides the inference encoder q_phi and the conditional decoder p_theta; ColaDiTModel parameterizes the block-causal latent prior p_psi. Diffusion is used to transport the latent prior (Eq. 2.1.4 of the paper), not to recover tokens.
- HuggingFace-native. ColaDiTModel and ColaTextVAEModel subclass transformers.PreTrainedModel and ship with matching PretrainedConfig classes, so from_pretrained / save_pretrained / AutoConfig all work out of the box.










