📢 Happy to introduce SiReRAG: our #ICLR2025 paper on RAG indexing!. Facilitating comprehensive knowledge synthesis on multihop reasoning, SiReRAG models both similarity and relatedness signals of a corpus. Code: Paper: (1/N)🧵.

RT @omarsar0: Enhancing RAG with Application-Aware Reasoning. Neat trick to improve RAG systems: give it the relevant knowledge and show it….

RT @rohanpaul_ai: Brilliant Paper. We need to evaluate reasoning steps separately for knowledge correctness and reasoning quality. LLMs….

RT @RyoKamoi: 📢 New paper!.FoVer enhances PRMs for step-level verification of LLM reasoning w/o human annotation 🚀.We synthesize training d….

RT @_philschmid: 100 Days After DeepSeek-R1. What have we learned? Where did we see success? What was most challenging? A Survey on Replica….

RT @ruizhang_nlp: As Vision Language Models treat images as tokens, high-resolution images create long sequences, similar to long-context c….

RT @ruizhang_nlp: This work is led by my first PhD student Yusen @YusenZhangNLP, who is graduating soon and actively seeking a postdoc posi….

RT @YusenZhangNLP: 🚀 How Far Are VLMs from Effective High-Resolution Image Understanding?.👉 We found: Still far. 🆕 Introducing HRScene Ben….

RT @Alibaba_Qwen: Introducing Qwen3! . We release and open-weight Qwen3, our latest large language models, including 2 MoE models and 6 den….

RT @Shiwei_Liu66: Our ICLR 2025 Workshop on Sparsity in LLMs (@sparseLLMs) kicks off with a talk by @DAlistarh on lossless (~1% perf drop)….

RT @SFResearch: 🔥 Phenomenal Day 1 of poster sessions at #ICLR25! ✨ . Attending tomorrow? Visit us at Booth #G03 to dive into how our groun….

RT @jasonwu0731: Heading to Singapore for #ICLR25 to present three papers: ReGenesis, BingoGuard, SiReRAG. We are investing heavily on Ag….

I will join #ICLR2025 virtually and my collaborators are presenting SiReRAG in person during Poster Session 1. We will be #61 at Hall 3 + Hall 2B. I am happy to discuss research on RAG, LLMs compression, and large reasoning models.🤝.

RT @sarkarssdas: Try GReaTerPrompt today to supercharge your prompts — whether you're using open-source models or API-based models! 👇.

RT @ruizhang_nlp: 📢 GreaterPrompt is Now Live!. We're excited to introduce GreaterPrompt, a unified, customizable, and high-performance ope….

RT @SFResearch: SiReRAG helps identify core bottlenecks in current #AISystems and develop elegant solutions that address them. By integrati….

RT @ruizhang_nlp: Excited to share SiReRAG! Our #ICLR2025 paper on improving RAG indexing for multihop reasoning. 🔍 SiReRAG combines simil….

This paper marks a wonderful internship experience at @SFResearch ! A huge shoutout to my amazing collaborators, Prafulla Kumar Choubey, @alexfabbri4 , Gabriel Bernadett-Shapiro, @ruizhang_nlp , Prasenjit Mitra, @CaimingXiong , and @jasonwu0731 !. (N/N).

Finally, SiReRAG showcases wide applicability on multihop QA across various retrieval methods. We show that SiReRAG significantly improves other non-indexing methods (e.g., reranking and iterative retrieval). (6/N) 🧵

SiReRAG delivers consistent improvement over state-of-the-art baselines on multihop datasets (an average 1.9% F1 gain), which shows the advantage of extensive knowledge synthesis. Baselines focus on either similarity (RAPTOR) or relatedness (HippoRAG and GraphRAG). (5/N) 🧵

SiReRAG stands for RAG indexing of similarity and relatedness: similarity via a recursive chunk-based tree, and relatedness through entity-based synthesis. Both fine-tuned LLMs and GPT-4o effectively extract entities and propositions. (4/N) 🧵