Transformer Architecture · 2025
Large Language Diffusion Models (LLaDA)
Shows a masked-diffusion (non-autoregressive) LLM can reach 8B scale and rival LLaMA3-8B, establishing parallel token generation as a viable alternative to left-to-right decoding.
Editorial record
Plain-language summary
Shows a masked-diffusion (non-autoregressive) LLM can reach 8B scale and rival LLaMA3-8B, establishing parallel token generation as a viable alternative to left-to-right decoding.
Knowledge graph
Relationships
Descendants
Depends onEvidence: Strongly supported
Attention Is All You Need
Builds on the transformer architecture lineage in the archive
freshness sweep 2026
Source record
Provenance
- Record ID
- P-611
- Record created
- 2026-07-13
- Last reviewed
- 2026-07-14
- Record version
- 2
- https://arxiv.org/abs/2502.09992
- arXiv:2502.09992
Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.