# KNN-LM Does Not Improve Open-ended Text Generation

arXiv: https://arxiv.org/abs/2305.14625 · HTML: https://mlsanity.com/paper/2305.14625/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Placed in RAG robustness and adaptive retrieval, one of 15 directions in Retrieval-augmented generation.
- Builds on 40 of its 48 references that are in the map.
- First posted on arXiv 24 May 2023.

## Placed in

- Retrieval-augmented generation / [RAG robustness and adaptive retrieval](https://mlsanity.com/rag/rag-robustness-and-adaptive-retrieval/)

## Builds on (most cited first, 30 of 40)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [The Curious Case of Neural Text Degeneration](https://mlsanity.com/paper/1904.09751/) (2019)
- [REALM: Retrieval-Augmented Language Model Pre-Training](https://mlsanity.com/paper/2002.08909/) (2020)
- [Improving language models by retrieving from trillions of tokens](https://mlsanity.com/paper/2112.04426/) (2021)
- [Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering](https://mlsanity.com/paper/2007.01282/) (2020)
- [Scaling Language Models: Methods, Analysis & Insights from Training Gopher](https://mlsanity.com/paper/2112.11446/) (2021)
- [Atlas: Few-shot Learning with Retrieval Augmented Language Models](https://mlsanity.com/paper/2208.03299/) (2022)
- [Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions](https://mlsanity.com/paper/2212.10509/) (2022)
- [Generalization through Memorization: Nearest Neighbor Language Models](https://mlsanity.com/paper/1911.00172/) (2019)
- [REPLUG: Retrieval-Augmented Black-Box Language Models](https://mlsanity.com/paper/2301.12652/) (2023)
- [Reading Wikipedia to Answer Open-Domain Questions](https://mlsanity.com/paper/1704.00051/) (2017)
- [Compressive Transformers for Long-Range Sequence Modelling](https://mlsanity.com/paper/1911.05507/) (2019)
- [In-Context Retrieval-Augmented Language Models](https://mlsanity.com/paper/2302.00083/) (2023)
- [Neural Text Generation with Unlikelihood Training](https://mlsanity.com/paper/1908.04319/) (2019)
- [MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontiers](https://mlsanity.com/paper/2102.01454/) (2021)
- [Evaluation of Text Generation: A Survey](https://mlsanity.com/paper/2006.14799/) (2020)
- [Why We Need New Evaluation Metrics for NLG](https://mlsanity.com/paper/1707.06875/) (2017)
- [Memorizing Transformers](https://mlsanity.com/paper/2203.08913/) (2022)
- [Nearest Neighbor Machine Translation](https://mlsanity.com/paper/2010.00710/) (2020)
- [Factuality Enhanced Language Models for Open-Ended Text Generation](https://mlsanity.com/paper/2206.04624/) (2022)
- [Hurdles to Progress in Long-form Question Answering](https://mlsanity.com/paper/2103.06332/) (2021)
- [Rethinking with Retrieval: Faithful Large Language Model Inference](https://mlsanity.com/paper/2301.00303/) (2022)
- [Unifying Human and Statistical Evaluation for Natural Language Generation](https://mlsanity.com/paper/1904.02792/) (2019)
- [How much do language models copy from their training data? Evaluating linguistic novelty in text generation using RAVEN](https://mlsanity.com/paper/2111.09509/) (2021)
- [Entity-level Factual Consistency of Abstractive Text Summarization](https://mlsanity.com/paper/2102.09130/) (2021)
- [Training Language Models with Memory Augmentation](https://mlsanity.com/paper/2205.12674/) (2022)
- [LongEval: Guidelines for Human Evaluation of Faithfulness in Long-form Summarization](https://mlsanity.com/paper/2301.13298/) (2023)
- [Efficient Nearest Neighbor Language Models](https://mlsanity.com/paper/2109.04212/) (2021)

## Built on by (most cited first, 10 of 10)

- [Ragas: Automated Evaluation of Retrieval Augmented Generation](https://mlsanity.com/paper/2309.15217/) (2023)
- [Understanding Retrieval Augmentation for Long-Form Question Answering](https://mlsanity.com/paper/2310.12150/) (2023)
- [PipeRAG: Fast Retrieval-Augmented Generation via Algorithm-System Co-design](https://mlsanity.com/paper/2403.05676/) (2024)
- [Nearest Neighbor Speculative Decoding for LLM Generation and Attribution](https://mlsanity.com/paper/2405.19325/) (2024)
- [Retrieval is Accurate Generation](https://mlsanity.com/paper/2402.17532/) (2024)
- [MoRSE: Bridging the Gap in Cybersecurity Expertise with Retrieval Augmented Generation](https://mlsanity.com/paper/2407.15748/) (2024)
- [Frustratingly Simple Retrieval Improves Challenging, Reasoning-Intensive Benchmarks](https://mlsanity.com/paper/2507.01297/) (2025)
- [DioR: Adaptive Cognitive Detection and Contextual Retrieval Optimization for Dynamic Retrieval-Augmented Generation](https://mlsanity.com/paper/2504.10198/) (2025)
- [CoCoLex: Confidence-guided Copy-based Decoding for Grounded Legal Text Generation](https://mlsanity.com/paper/2508.05534/) (2025)
- [Training-Free Token-Level Steering for LLM Personalized Co-Writing](https://mlsanity.com/paper/2608.06069/) (2026)

## Abstract (arXiv)

In this paper, we study the generation quality of interpolation-based retrieval-augmented language models (LMs). These methods, best exemplified by the KNN-LM, interpolate the LM's predicted distribution of the next word with a distribution formed from the most relevant retrievals for a given prefix. While the KNN-LM and related methods yield impressive decreases in perplexity, we discover that they do not exhibit corresponding improvements in open-ended generation quality, as measured by both automatic evaluation metrics (e.g., MAUVE) and human evaluations. Digging deeper, we find that interpolating with a retrieval distribution actually increases perplexity compared to a baseline Transformer LM for the majority of tokens in the WikiText-103 test set, even though the overall perplexity is lower due to a smaller number of tokens for which perplexity dramatically decreases after interpolation. However, when decoding a long sequence at inference time, significant improvements on this smaller subset of tokens are washed out by slightly worse predictions on most tokens. Furthermore, we discover that the entropy of the retrieval distribution increases faster than that of the base LM as the generated sequence becomes longer, which indicates that retrieval is less reliable when using model-generated text as queries (i.e., is subject to exposure bias). We hope that our analysis spurs future work on improved decoding algorithms and interpolation strategies for retrieval-augmented language models.