NeoMME: 255x smaller multimodal embeddings while keeping >95% nDCG@10
antoine_chaffin · x · 2026-11-12
Working with multimodal embedding model NeoMME, researchers found that token pooling + asymmetric quantization — with no pooling-aware training at all — shrinks document embeddings 255x while retaining >95% of nDCG@10. They argue dedicated training could push further, since many image patches are near-empty.
Tooling: Sentence Transformers v6.3 adds token pooling-aware training to build multi-vector models that preserve retrieval quality with fewer document vectors; the open-source late-interaction-kernels repo provides fused Triton kernels for late-interaction (MaxSim) scoring as used by ColBERT/ColPali, with docs covering tiling, online max, the backward pass, and benchmarks.
More from Research
- PhyRubric: rubric-based benchmark diagnoses physical commonsense in text-to-video — Huiyi Chen · 2026-11-12
- OpenAI auto-researchers publish Yang-Mills validation, mass gap still unsolved — IgorCarron · 2026-11-12
- Economist: 3 reasons letting AI write your paper abstract is a mistake — Afinetheorem · 2026-11-12
- CMU and JetBrains study: doubling AI token spend yields only ~10% more commits — rohanpaul_ai · 2026-11-12
- ADJ: modeling LLM jailbreaking as a Hegelian-dialectic Stackelberg game between two LLMs — chaumian · 2026-11-12
- AI-Designed T-Cell Receptors Boost Cancer Killing in Lab and Tumor Control in Mice — WmHaseltine · 2026-11-12