Colbert
PulseAugur coverage of Colbert — every cluster mentioning Colbert across labs, papers, and developer communities, ranked by signal.
2 day(s) with sentiment data
-
Perplexity releases contextual embedding model for RAG pipelines
Perplexity Research and turbopuffer have released pplx-embed-v2-context-9b-preview, a new contextual embedding model designed for retrieval-augmented generation (RAG) pipelines. This model differs from traditional appro…
-
LLM-derived weights enhance psychometric model for educational assessment reliability
Researchers have developed a new Rater Ising-Potts model that leverages Large Language Model (LLM) embeddings to assess the reliability of educational assessments. This model focuses on pairwise agreement between rating…
-
EigenLI framework offers spectral approximation for late-interaction models
Researchers have developed EigenLI, a novel framework for approximating late-interaction models in information retrieval. This method leverages the intrinsic low-rank structure of document token embeddings to compress r…
-
University of Aveiro team details BioASQ 14B biomedical QA system
The BIT.UA team from the University of Aveiro has detailed their participation in the BioASQ 14B challenge, focusing on biomedical question answering. They implemented a modular system that refactored both retrieval and…
-
MoganColBERT-TR: New Turkish Multi-Vector Retrieval Model Unveiled
Researchers have developed MoganColBERT-TR, a new multi-vector retrieval model specifically designed for the Turkish language. This model builds upon a previously trained ModernBERT encoder and adapts it to the ColBERT …
-
Hugging Face details training multi-vector embedding models
Hugging Face has released a blog post detailing how to train and fine-tune multi-vector embedding models using the sentence-transformers library. This approach, inspired by ColBERT-style late interaction retrieval, allo…
-
Hugging Face introduces MultiVectorEncoder for advanced retrieval
Hugging Face has introduced MultiVectorEncoder, a new tool within its sentence-transformers library that enables the use of multi-vector embedding models. These models, inspired by the ColBERT architecture, process text…
-
ColGraphRAG enhances multimodal QA with late-interaction image retrieval
Researchers have introduced ColGraphRAG, a novel approach to multimodal question answering that enhances evidence retrieval by focusing on late-interaction scoring for graph-linked images. This method aims to improve ac…
-
PLAID-PRF enhances dense retrieval with centroid-aware pseudo-relevance feedback
Researchers have introduced PLAID-PRF, a novel method for enhancing multi-vector dense retrieval models like ColBERT. This technique leverages centroid-like tokens within the PLAID framework to perform Pseudo-Relevance …
-
Qdrant cuts RAG token costs by 67% with native ColBERT reranking
Qdrant has introduced a native ColBERT reranking feature that significantly reduces token costs for Retrieval-Augmented Generation (RAG) systems. This new capability allows Qdrant to perform token-to-token comparisons d…
-
New theory explains late-interaction retrieval models, introduces Signed MaxSim
Researchers have theoretically quantified the representational power of late-interaction retrieval models, specifically those using the MaxSim similarity function. The study demonstrates that MaxSim can precisely replic…
-
TileMaxSim kernel boosts GPU retrieval model speed by 220x
Researchers have developed TileMaxSim, a new IO-aware kernel for GPUs designed to significantly accelerate the MaxSim scoring process used in multi-vector retrieval models like ColBERT. Existing implementations are inef…
-
HAKARI-Bench offers lightweight evaluation for retrieval models · 2 sources tracked
Researchers have introduced HAKARI-Bench, a lightweight benchmark designed to streamline the evaluation of retrieval architectures and efficiency settings for retrieval-augmented generation and semantic search. This new…
-
AI Search Needs Tensors, Not Just Vectors, for Production
Production AI systems require more than basic vector search, which struggles to integrate structured attributes, business rules, personalization, and ML ranking models. Tensors offer a solution by allowing multi-dimensi…
-
Dr-DCI framework scales agentic search with dynamic workspace expansion
Researchers have developed Dr-DCI, a novel framework designed to enhance agentic search capabilities over large corpora. This system dynamically expands a local workspace by retrieving relevant documents, allowing agent…
-
ColBERTSaR shrinks ColBERT indexes by 70% using quantization
Researchers have developed ColBERTSaR, a novel method for sparsifying ColBERT indexes using product quantization. This technique significantly reduces the index size, making it 50-70% smaller than previous implementatio…
-
New retrieval method replaces K-means with sparse coding for faster, more accurate results
Researchers have introduced Single-stage Sparse Retrieval (SSR), a new method for efficient multi-vector retrieval that bypasses traditional K-means clustering. SSR utilizes Sparse Autoencoders to create high-dimensiona…
-
New GPU Kernel Speeds Up AI Retrieval Tasks, Cuts Memory Use
Researchers have developed Flash-MaxSim, a novel IO-aware fused GPU kernel designed to optimize late-interaction retrieval scoring. This new kernel computes the same scores as standard implementations but avoids materia…
-
ProtoCol model enhances protein homolog search using late interaction
Researchers have developed ProtoCol, a novel model designed to improve protein homolog search, particularly in challenging "twilight zone" scenarios where sequence similarity is low. ProtoCol utilizes residue-level embe…
-
New Multilingual ColBERT Model Excels in Clinical Text Analysis
Researchers have developed ClinicalEncoder26AM, a multilingual Diagnosable ColBERT model specifically designed for clinical and biomedical texts. This model aligns token-level semantics with a clinical latent space, Cli…