Metraipolis › archive › dc_10ba5ed0cb28
Telescopic Language Models — arXiv:2609.35769
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.35769
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T21:03:58.246Z
- SHA-256 of the archived text
2432a18653721d54f18860ecd5c2fc3398e090dd69565d4da5fd980a100dc428- Stable URL of this page
https://metraipolis.com/documento/dc_10ba5ed0cb28
How to cite
Telescopic Language Models — arXiv:2609.35769 — arXiv (en), https://arxiv.org/abs/2609.35769 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_10ba5ed0cb28 (Metraipolis archive dc_10ba5ed0cb28, sha256 2432a18653721d54f18860ecd5c2fc3398e090dd69565d4da5fd980a100dc428)
BibTeX
@misc{dc10ba5ed0cb28,
title = {Telescopic Language Models — arXiv:2609.35769},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.35769},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_10ba5ed0cb28; sha256 2432a18653721d54f18860ecd5c2fc3398e090dd69565d4da5fd980a100dc428},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35769. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
One deployed language model must often serve many compute budgets, yet serving each budget still means a separate training or compression run per point. We train a Telescopic Language Model (TLM) to be that continuum: a nested-capacity Transformer supervised by stochastic prefix supervision with a full anchor. At every step, one randomly truncated prefix of the capacity axis is trained against the full next-token target, alongside one full-capacity pass, so the trained artifact is a valid language model at every depth. Two forward-backward passes per step, no architectural change, nothing extra at inference. Fixed-exit suites such as Matryoshka Language Model Suites (MLMS) occupy one point in this design space, and the point has a cost: supervising only a few fixed exits leaves the nested model at chance level everywhere else (perplexity 10^2-10^5 in our baselines). On a 200M proxy suite (20B FineWeb-Edu tokens, identical data stream for all methods), a single TLM run is a valid language model at every one of its twenty layer prefixes, in perplexity and on perplexity-sensitive downstream tasks, reducing the area under the quality-budget curve by 43-44% relative to the fixed-exit suites while matching them at full capacity, at ~12% lower GPU cost per run. The prefix sampling density is a dial: concentrating it on a few depths recovers fixed-exit quality there at the price of the continuum, so the operating points become a training-time choice rather than an architectural one. These results indicate that the training objective, not the nesting itself, is what makes a model elastic. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.35769 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T20:29:03.452Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).