Metraipolis › archive › dc_f560f66e5477
HydraLLM: Adaptive Width Slicing for Elastic Text and Vision-Language Models — arXiv:2610.23455
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.23455
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T08:43:50.968Z
- SHA-256 of the archived text
ad1aa11cb9ff8d5227b5ae150e436922bc4de0640d3a96404e7a1d77141f6c87- Stable URL of this page
https://metraipolis.com/documento/dc_f560f66e5477
How to cite
HydraLLM: Adaptive Width Slicing for Elastic Text and Vision-Language Models — arXiv:2610.23455 — arXiv (en), https://arxiv.org/abs/2610.23455 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f560f66e5477 (Metraipolis archive dc_f560f66e5477, sha256 ad1aa11cb9ff8d5227b5ae150e436922bc4de0640d3a96404e7a1d77141f6c87)
BibTeX
@misc{dcf560f66e5477,
title = {HydraLLM: Adaptive Width Slicing for Elastic Text and Vision-Language Models — arXiv:2610.23455},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.23455},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f560f66e5477; sha256 ad1aa11cb9ff8d5227b5ae150e436922bc4de0640d3a96404e7a1d77141f6c87},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.23455. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
LLMs and LVLMs are increasingly deployed across heterogeneous hardware, ranging from data-center GPUs to resource-constrained edge devices. Their fixed-size architectures limit the ability to reuse a single model across deployment settings with different compute, latency, and model-size targets. We introduce HydraLLM, an elastic MQA-based architecture. Its number of active attention heads determines the effective widths of the embeddings, attention projections, feed-forward layers, and the output head, forming a nested family of subnetworks within a single parameter set. We introduce a static-to-elastic conversion pipeline that, starting from a pre-trained model checkpoint, includes importance-driven weight reordering and stochastic subnetwork training with knowledge distillation. We extend this principle to multimodal inference through an elastic projection module whose output width follows the active text width. HydraLLM covers a wider range of active parameter configurations than FFN-only elastic architectures. With 75% of heads active, HydraLLM achieves 2.2 percentage points higher TriviaQA accuracy than a state-of-the-art nested baseline configuration which has 17.25% more parameters. With 50% of heads active, our multimodal model achieves a CLIPScore comparable to that of the 7$\times$ larger Gemma 3 4B model. With only 25% of heads active, HydraLLM increases token-generation throughput by 75.3% over the full-width configuration. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.23455 — 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 2026-10-19T08:39:24.870Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).