Tag
Introduces SaMer, an object-aware token merging framework that compresses image-side tokens for vision-language retrieval while preserving object-level evidence, achieving significant storage reduction and improved retrieval performance.
TACHIOM, a multivector retrieval system with token-aware clustering and hierarchical indexing, has been integrated into the PyLate ecosystem. It achieves up to 247x faster clustering and 9.8x retrieval speedup over state-of-the-art systems while maintaining comparable effectiveness.
Turbovec, based on Google's TurboQuant algorithm, compresses 10 million vectors from 31GB to 4GB, with search speed 12-20% faster than FAISS, supports filtered search, and offers a Rust implementation with a Python package.
This paper presents a single-stage sparse coding method using unsupervised sparse autoencoders and natural inverted indexing to accelerate multi-vector retrieval, outperforming traditional k-means based approaches.
Introduces SimpleMem, an efficient memory framework for LLM agents that uses semantic lossless compression to improve accuracy and reduce token consumption, achieving 26.4% F1 improvement and up to 30x reduction in inference-time token usage.