{"ID":2885191,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2508.05305","arxiv_id":"2508.05305","title":"SONAR-LLM: Autoregressive Transformer that Thinks in Sentence Embeddings and Speaks in Tokens","abstract":"The recently proposed Large Concept Model (LCM) generates text by predicting a sequence of sentence-level embeddings and training with either mean-squared error or diffusion objectives. We present SONAR-LLM, a decoder-only transformer that \"thinks\" in the same continuous SONAR embedding space, yet is supervised through token-level cross-entropy propagated via the frozen SONAR decoder. This hybrid objective retains the semantic abstraction of LCM while eliminating its diffusion sampler and restoring a likelihood-based training signal. Across model sizes from 39M to 1.3B parameters, SONAR-LLM attains competitive generation quality. We report scaling trends, ablations, benchmark results, and release the complete training code and all pretrained checkpoints to foster reproducibility and future research.","short_abstract":"The recently proposed Large Concept Model (LCM) generates text by predicting a sequence of sentence-level embeddings and training with either mean-squared error or diffusion objectives. We present SONAR-LLM, a decoder-only transformer that \"thinks\" in the same continuous SONAR embedding space, yet is supervised through...","url_abs":"https://arxiv.org/abs/2508.05305","url_pdf":"https://arxiv.org/pdf/2508.05305v2","authors":"[\"Nikita Dragunov\",\"Temurbek Rahmatullaev\",\"Elizaveta Goncharova\",\"Nikita Kurdiukov\",\"Aysel Mirzoeva\",\"Anna Borisiuk\",\"Andrey Kuznetsov\",\"Anton Razzhigaev\"]","published":"2025-08-07T12:03:44Z","proceeding":"cs.CL","tasks":"[\"cs.CL\"]","methods":"[\"Diffusion Model\",\"Transformer\",\"Large Language Model\"]","has_code":false}