Our paper on the core architecture behind tusKANNy is out.
Two libraries, one idea: separate the data from the index. 🧵
#VectorSearch #NeuralIR #RustLangView quoted post on X
Franco Maria Nardini
@fmnardini
Research Director @ ISTI-CNR, Pisa, Italy. Information Retrieval, Vector Search, Machine/Deep Learning, Efficiency.
1.6K Following 804 Followers 527 Posts
Latest posts 5
TACHIOM is now in PyLate: lightning-fast multi-vector indexing and search directly on CPU!
arxiv.org/abs/2604.28142
Joint work with @SilvioMartinico, @cosimorulli1, @rventurini_.
Thanks, @antoine_chaffin, and the PyLate team, for the support with the integration!
arxiv.org/abs/2604.28142
Joint work with @SilvioMartinico, @cosimorulli1, @rventurini_.
Thanks, @antoine_chaffin, and the PyLate team, for the support with the integration!
arxiv.org Efficient Multivector Retrieval with Token-Aware Clustering and... Multivector retrieval models achieve state-of-the-art effectiveness through fine-grained token-level representations, but their deployment incurs substantial computational and memory costs.... Whether you are GPU poor or GPU rich, today's release of PyLate has something for you!
GPU maxxers: MaxSim kernels greatly speed up training while lowering the memory requirements
CPU enjoyers: TACHIOM enables lightning fast multi-vector indexing and search directly on CPUView quoted post on X
Interested in how to efficiently re-rank with cross-encoders? Join us today (Efficiency session, 10:30 - 12:30) to hear our talk on "Efficient Re-ranking with Cross-encoders via Early Exit" #SIGIR2025
Six accepted contributions that span several topics, from efficient neural IR to conversational search, counterfactual user modeling, and resource-efficient AI. #SIGIR2025
🚀 Exciting news from the #HPCLab @Cnr_Isti !
We are proud to share that our researchers will present six papers at @SIGIRConf 2025, the flagship conference in #InformationRetrieval and Search Technologies, which will take place from July 13 to 17 in Padua, Italy.View quoted post on X
"Neural Network Compression using Binarization and Few Full-precision Weights." Joint work with @cosimorulli1, @tranisalvatore, and @rventurini_. Elsevier Information Systems. Just Accepted.
doi.org/10.1016/j.ins.…
doi.org/10.1016/j.ins.…