arXiv:2605. 04344v2 Announce Type: replace-cross Abstract: This paper develops a statistical theory of extrapolation for large language models, by reinterpreting them through pre-post-additive noise models.
arXiv:2512. 20757v2 Announce Type: replace-cross Abstract: Tokenizers provide the fundamental basis through which text is represented and processed by language models (LMs).
The paper explores structural transfer, where models are first trained on non-language data such as music, probabilistic grammars, and cellular automata to induce priors for natural language tasks. This pretraining acts as a weight initialization for multilingual language modeling and leads to lower next-token prediction loss and smaller weight shifts during subsequent language training. However, the improved loss does not consistently translate into better downstream linguistic performance, and the efficiency of non-language data is lower than that of additional language data.
The paper introduces a training strategy for cascaded simultaneous speech translation that allows the system to dynamically decide how much of the source prefix to translate. By fine‑tuning a large language model (Qwen3‑8B) on stable prefixes—pairs of source prefixes and the longest shared translation with the full sentence—the authors enable contextual read‑write decisions beyond fixed wait‑k or target‑suffix deletion. Experiments on English‑to‑German, Japanese, and Chinese demonstrate that stable prefixes improve the quality‑latency tradeoff across various test sets.
arXiv:2606. 16246v1 Announce Type: cross Abstract: As AI labs approach a data ceiling where compute capacity outpaces the rate of new high-quality text generation, language model pretraining is shifting toward a data-constrained, compute-abundant regime that demands productive multi-epoch training on fixed corpora.
Manacá-1B is a 1.72‑billion‑parameter, open decoder‑only language model trained from scratch for Brazilian Portuguese, released with a fully containerized, reproducible training pipeline and complete logs. The authors evaluate it against nine open baselines on four Portuguese benchmarks, reporting standard errors and paired significance tests, and find that Manacá-1B outperforms smaller models on LAMBADA‑PT while remaining competitive on commonsense completion. They also uncover a tokenizer‑related evaluation pitfall that can drastically lower accuracy and provide a simple fix, releasing all code, logs, and corrected tokenizer for full reproducibility.
The paper introduces TokenAdapt, a model‑agnostic tokenizer transplantation method that uses a hybrid heuristic to initialize new token embeddings, and a novel pre‑tokenization learning approach for multi‑word Supertokens to improve compression. TokenAdapt combines local subword decomposition and global semantic similarity to preserve semantics while reducing retraining needs. Empirical results show that TokenAdapt outperforms existing baselines such as Transtokenizer and ReTok, achieving lower perplexity ratios and significant compression gains.
The paper investigates training-time data augmentation as a regularizer for autoregressive language model pretraining in data‑constrained, compute‑abundant settings. It introduces three orthogonal augmentation categories—token‑level noise, sequence permutations, and target offset prediction—and shows through systematic ablations that each category delays overfitting and reduces validation loss, with random token replacement performing best individually. Combining augmentation categories further lowers the minimum validation loss, demonstrating that such augmentations mitigate data inefficiency in autoregressive pretraining.
arXiv:2608. 15448v1 Announce Type: cross Abstract: Large language models increasingly rely on sampling as a driver of their own improvement, making the fidelity of their learned distributions more critical than ever.
The paper introduces a new language modeling approach that combines the benefits of Transformers and recurrent models by using balanced-tree recursive operators for autoregressive prediction. This method achieves logarithmic depth and linear runtime, allowing all prefix representations to be computed efficiently. Experiments show strong length extrapolation and performance close to ALiBi-based Transformers, suggesting it could serve as a viable alternative architecture for language modeling.
The study investigates how textual neural models degrade when inputs contain noise such as typos, OCR errors, or dropped words. It finds that model performance decline is largely consistent across architectures under word‑level noise but diverges under character‑level noise, a difference attributed to tokenization rather than architecture. By applying a short contrastive training recipe, diverse encoders converge to a common robustness curve, enabling prediction of a model’s noise resilience and the ability to enhance robustness at specific noise scales through targeted training.
arXiv:2506. 07673v2 Announce Type: replace Abstract: Large language model (LLM) evaluation is increasingly costly, prompting interest in methods that speed up evaluation by shrinking benchmark datasets.
The article presents a mathematical framework for large language models (LLMs), detailing how text sequences are encoded into tokens, how next‑token prediction architectures are defined, and how these models are trained and deployed for tasks such as summarization, recommendation, software writing, and quantitative problem solving. It emphasizes that the framework relies on basic concepts from information theory, probability, and optimization, yet captures the complex algorithmic structure responsible for LLMs’ empirical successes. The authors argue that this formalism enables the study of accuracy, efficiency, and robustness, and points toward new methodological developments.