Query Expension for Better Query Embedding using LLMs
-
Updated
Feb 18, 2025 - Python
Query Expension for Better Query Embedding using LLMs
Code and models for the paper "Questions Are All You Need to Train a Dense Passage Retriever (TACL 2023)"
SPRINT Toolkit helps you evaluate diverse neural sparse models easily using a single click on any IR dataset.
Swap one RAG pipeline stage, hold the rest fixed, and measure whether the gain survives composition. The framework behind "Retrieval Augmented Generation Compositionality": 450 configurations, eight benchmarks, corpora from 5K to 200K documents.
Evaluation of BEIR Datasets using ColBERT retrieval model
LoRA fine-tuning of bi-encoder retrievers with hard negatives and cross-encoder distillation, evaluated on NFCorpus.
A decentralized cooperative memory & research layer for AI agents — collectively and cooperatively learning and advancing as a community.
A genral RAG Search chatbot, with SoTA RAG techniques such as HyDE, Hybrid retrieval with BM25 + RRF and Cross encoder reranking. Evaluated on the BEIR scifact dataset and compared all the different pipelines i tried along the way
Local search over your Markdown notes for Claude Code and Cursor, with re-runnable BEIR benchmarks.
Reproducible retrieval quality benchmark methods and curated result bundles for Cortrix semantic storage.
FlashRank-Pro: 149M-parameter ModernBERT cross-encoder reranker for RAG. Apache 2.0. Measured 0.3314 five-set BEIR-style NDCG@10 (+22.6% vs run-init). CPU-friendly, 8K context.
Vector vs keyword search benchmark: BM25 in Elasticsearch vs dense FAISS + MiniLM vs hybrid RRF retrieval on BEIR SciFact, behind one Retriever protocol, scored on Recall, Precision, MRR and nDCG. Python.
Retrieval pipelines for RAG measured on public relevance judgements: BM25, dense, RRF hybrid, reranking and chunking on BEIR, with paired significance tests and run records that CI re-executes.
Zero-label BM25 hyperparameter adaptation — O (1) lookup from corpus statistics, +18% NDCG@10 on BEIR, Rust-accelerated
A modular Information Retrieval (IR) engine and Streamlit dashboard benchmarking fast hybrid search (BM25 + Dense Vectors) against deep generative pipelines (HyDE + Cross-Encoder Rerankers).
Research-grade hybrid retrieval API — BM25 + FAISS + CDF calibration + entropy-weighted fusion + cross-encoder reranking. Benchmarked on BEIR SciFact with bootstrap significance tests.
Rigorous retrieval evaluation over BEIR FiQA-2018: hand-written nDCG/recall/MRR/MAP validated against pytrec_eval to 1e-6, hybrid BM25+dense fusion, cross-encoder reranking, paired-bootstrap significance, and a single-use held-out test split.
Banc d'évaluation du retrieval sur BEIR SciFact : métriques codées à la main, campagnes avec prédiction écrite avant mesure, runs suivis dans MLflow.
Hybrid retrieval built to be measured, not believed — BM25 + dense + HNSW + fusion + re-ranking in pure numpy. Hand-rolled BM25 reproduces published BEIR baselines to 0.0066 MAD.
BM25 and dense retrieval in Go, where each component is accepted against a number Pyserini publishes for BEIR and every result reproduces from public commands and data.
To associate your repository with the beir topic, visit your repo's landing page and select "manage topics."