Adrien Deschenaux, Nadezhda Ilieva, Justin Deschenaux, Manuel Madeira, Dorina Thanou, Aurélien Lucchi, Antonio Orvieto, Caglar Gulcehre · 2026 · arXiv
DOI: 10.48550/arXiv.2610.30068 · open-access PDF
Recentes modelos de linguagem de difusão (DLMs) usam auto-condicionamento (SC), ou Loopholing, para melhorar seu desempenho. Enquanto SC é tradicionalmente visto como um heurístico auxiliar, seu sucesso empírico levanta a questão: pode um modelo de geração contínua ser baseado puramente em auto-condicionamento? Nós respondemos afirmativamente com LAW ("Loopholing All the Way"), que propaga o Embedding esperado sob a previsão do denoiser através de etapas, decodificando os tokens uma vez, no final.
Machine translation (Meta M2M100); the English original follows.
Recent Diffusion Language Models (DLMs) use Self-Conditioning (SC), or Loopholing, to improve their performance. While SC is traditionally seen as an auxiliary heuristic, its empirical success raises the question: can a continuous generative model be based purely on Self-Conditioning? We answer affirmatively with LAW (``Loopholing All the Way''), which propagates the Expected Embedding under the denoiser's prediction across steps, decoding the tokens once, at the end. On Sudoku, LAW outperforms previous Discrete and Continuous Diffusion Models, including Loopholing Discrete Diffusion. On Maze,
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.30068, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange