Very excited to announce new work with Charlie O'Neill (& John F. Wu (@jwuphysics.bsky.social) , Kartheik Iyer, JHU CLSP) training sparse autoencoders over embeddings, discovering thousands of human-interpretable features from scientific papers, and using SAE interventions to improve semantic search

I am thrilled to announce the release of a long-time teased project built with Raphaël Sourty Introducing PyLate, a flexible library to train and experiment with ColBERT models! It's built on the widely-used sentence-transformers framework, making it accessible and efficient!

This paper looks like a banger 👀 Why isn't there any promo/hype about it on social? (cc Luca Soldaini 🎀)

Sentence Transformers v3.1 is out! Featuring a hard negatives mining utility to get better models out of your data, a new strong loss function, training with streaming datasets, custom modules, bug fixes, small additions and docs changes. Release notes: github.com/UKPLab/sentenc… 🧵

Promptriever: Instruction-Trained Retrievers Can Be Prompted Like Language Models Proposes a retrieval model that follows natural language instructions, enabling more flexible, and user-friendly search experiences. 📝arxiv.org/abs/2409.11136 👨🏽‍💻github.com/orionw/promptr…

Promptriever: Control your retrieval model's behavior with natural language! Thank you to Orion Weller the **awesome** work at Samaya AI this summer (and for showing me the ropes for IR research :-) ).

Traditional information retrieval (IR) systems struggle to adapt to user-specific instructions dynamically. Promptriever is the first IR model that can be prompted like a language model. Refine your searches with simple natural language prompts! This new approach makes IR much

hehehe i got a sneak peek from Orion Weller about this work last week 🫣 super excited to see more retrieval models designed to be able to follow complex instructions!!!

✨Promptriever✨ Promptriever 是第一个可以像 LLM 一样接受自然语言指令的双编码器检索模型，它可以根据每个查询的具体指令动态调整相关性定义，而不是使用固定的相关性标准。 研究方法： - 作者基于 MS MARCO 数据集创建了一个包含约 50 万实例的新指令训练数据集。 -