From 9910361fb84e86642157be0e2747a6c495c60fa7 Mon Sep 17 00:00:00 2001 From: Simon Suo Date: Sat, 6 Apr 2024 18:56:17 -0700 Subject: [PATCH 1/2] wip --- README.md | 2 ++ papers/adaptive-rag.md | 10 ++++++++++ 2 files changed, 12 insertions(+) create mode 100644 papers/adaptive-rag.md diff --git a/README.md b/README.md index 0deff8a..be254a7 100644 --- a/README.md +++ b/README.md @@ -46,6 +46,8 @@ - [ActiveRAG: Revealing the Treasures of Knowledge via Active Learning](./papers/active_rag.md) - Enhances RAG by active learning to deepen LLMs' understanding of external knowledge through innovative Knowledge Construction and Cognitive Nexus mechanisms. (Xu, Zhipeng, et al. 2024) +- [Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity](./papers/adaptive-rag.md) - The paper proposes an adaptive question-answering framework that dynamically selects the most suitable strategy for retrieval-augmented large language models based on the complexity of the query, using a classifier trained on automatically collected labels. (Jeong, Soyeong, et al. 2024) + ### RAG vs Finetuning - [RAG vs Fine-tuning: Pipelines, Tradeoffs, and a Case Study on Agriculture](./papers/rag_finetuning_agriculture.md) - RAG vs Fine-tuning case study on agriculture domain datasets. (Gupta, Aman, et al. 2024) diff --git a/papers/adaptive-rag.md b/papers/adaptive-rag.md new file mode 100644 index 0000000..fbffc9e --- /dev/null +++ b/papers/adaptive-rag.md @@ -0,0 +1,10 @@ +# [Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity](https://arxiv.org/abs/2403.14403) + +## Contributions + +## Results + +## Insights + +## Resources +- [LlamaIndex <> Mistral cookbook on Adaptive-RAG](https://github.com/mistralai/cookbook/blob/7bf0aae46ab8c763efb7800f352bcbfd8aceb8fb/third_party/LlamaIndex/Adaptive_RAG.ipynb) From 75fc8f1e3b292d79db14eecd5bb91d409dfd273f Mon Sep 17 00:00:00 2001 From: Simon Suo Date: Sat, 6 Apr 2024 19:00:18 -0700 Subject: [PATCH 2/2] wip --- papers/adaptive-rag.md | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/papers/adaptive-rag.md b/papers/adaptive-rag.md index fbffc9e..3bb9fcc 100644 --- a/papers/adaptive-rag.md +++ b/papers/adaptive-rag.md @@ -1,10 +1,19 @@ # [Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity](https://arxiv.org/abs/2403.14403) +> This paper card is generated with Claude 3 Opus ## Contributions +- Proposes Adaptive-RAG, a framework that dynamically adjusts retrieval-augmented LLM strategies based on query complexity, spanning from non-retrieval for straightforward queries to multi-step approaches for complex queries. +- Introduces a classifier to determine query complexity, which is trained on automatically collected data from model prediction outcomes and dataset biases. ## Results +- Adaptive-RAG significantly improves overall accuracy and efficiency compared to existing one-size-fits-all approaches on a collection of open-domain QA datasets covering diverse query complexities. +- The classifier effectively categorizes queries into different complexity levels, contributing to the selection of the most suitable retrieval-augmented LLM strategy. ## Insights +- Real-world queries have varying complexities, and existing retrieval-augmented generation approaches tend to be overly simple or complex. +- Adapting retrieval-augmented LLMs to the assessed query complexity enables the utilization of the most suitable approach tailored to each query. +- The query complexity classifier plays a crucial role in the adaptive framework, and there is still room for improvement in its architecture and training data. ## Resources - [LlamaIndex <> Mistral cookbook on Adaptive-RAG](https://github.com/mistralai/cookbook/blob/7bf0aae46ab8c763efb7800f352bcbfd8aceb8fb/third_party/LlamaIndex/Adaptive_RAG.ipynb) +- Official Code: https://github.com/starsuzi/Adaptive-RAG \ No newline at end of file