SlideOn hits the storage-efficiency frontier via Matryoshka training and vector pruning
IgorCarron · x · 2026-10-19
SlideOn reaches the storage-efficiency frontier by document vector pruning, with Matryoshka-style training and pruning budgets as the secret sauce. Charts show its d-128 to d-full variants dominate sparse baselines like SPLADE-v3 and OpenSearch on BEIR-15 vs NNZ, with quality flattening after d-256 (250 nonzero features/doc).
More from Research
- GlobeBench: 1,632 tasks show LLMs still struggle as text-based world models — PatronusAI · 2026-10-20
- Adobe's OmniXRef unifies text, image, and video references for video editing — adobe-research · 2026-10-20
- Stanford's Librarian Agents use RL to curate workspaces, beating retrieval by 33 points — StanfordUniversity · 2026-10-20
- Probe repo: GPT-4, with pre-GRPO cutoff, predicts DeepSeekMath's GRPO from anonymized text — CatAstro_Piyush · 2026-10-20
- Mistral launches Timestral time-series foundation models, zero-shot SOTA on GIFT-Eval — victormustar · 2026-10-20
- MD simulations reveal striking hydration dynamics in daraxonrasib tri-complex pockets — CatAstro_Piyush · 2026-10-20