RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding

Guanzheng Chen, Qilong Feng, Jinjie Ni, Xin Li, Michael Qizhe Shieh
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:8093-8107, 2025.

Abstract

The emergence of long-context large language models (LLMs) offers a promising alternative to traditional retrieval-augmented generation (RAG) for processing extensive documents. However, the computational overhead of long-context inference presents significant efficiency challenges. While Speculative Decoding (SD) traditionally accelerates inference using smaller draft models, its effectiveness diminishes substantially in long-context scenarios due to memory-bound KV cache operations. We introduce Retrieval-Augmented Speculative Decoding (RAPID), which leverages RAG for both accelerating and enhancing generation quality in long-context inference. RAPID introduces the RAG drafter—a draft LLM operating on shortened retrieval contexts—to speculate on the generation of long-context target LLMs. Our approach enables a new paradigm where same-scale or even larger LLMs can serve as RAG drafters while maintaining computational efficiency. To fully leverage the potentially superior capabilities from stronger RAG drafters, we develop an inference-time knowledge transfer that enriches the target distribution by RAG. Extensive experiments on the LLaMA-3.1 and Qwen2.5 backbones demonstrate that RAPID effectively integrates the strengths of both RAG and long-context LLMs, achieving significant performance improvements (e.g., from 39.33 to 42.83 on InfiniteBench for LLaMA-3.1-8B) with more than 2$\times$ speedups for long-context inference. Our analyses also reveal the robustness of RAPID across various context lengths and retrieval quality.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-chen25s, title = {{RAPID}: Long-Context Inference with Retrieval-Augmented Speculative Decoding}, author = {Chen, Guanzheng and Feng, Qilong and Ni, Jinjie and Li, Xin and Shieh, Michael Qizhe}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {8093--8107}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/chen25s/chen25s.pdf}, url = {https://proceedings.mlr.press/v267/chen25s.html}, abstract = {The emergence of long-context large language models (LLMs) offers a promising alternative to traditional retrieval-augmented generation (RAG) for processing extensive documents. However, the computational overhead of long-context inference presents significant efficiency challenges. While Speculative Decoding (SD) traditionally accelerates inference using smaller draft models, its effectiveness diminishes substantially in long-context scenarios due to memory-bound KV cache operations. We introduce Retrieval-Augmented Speculative Decoding (RAPID), which leverages RAG for both accelerating and enhancing generation quality in long-context inference. RAPID introduces the RAG drafter—a draft LLM operating on shortened retrieval contexts—to speculate on the generation of long-context target LLMs. Our approach enables a new paradigm where same-scale or even larger LLMs can serve as RAG drafters while maintaining computational efficiency. To fully leverage the potentially superior capabilities from stronger RAG drafters, we develop an inference-time knowledge transfer that enriches the target distribution by RAG. Extensive experiments on the LLaMA-3.1 and Qwen2.5 backbones demonstrate that RAPID effectively integrates the strengths of both RAG and long-context LLMs, achieving significant performance improvements (e.g., from 39.33 to 42.83 on InfiniteBench for LLaMA-3.1-8B) with more than 2$\times$ speedups for long-context inference. Our analyses also reveal the robustness of RAPID across various context lengths and retrieval quality.} }
Endnote
%0 Conference Paper %T RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding %A Guanzheng Chen %A Qilong Feng %A Jinjie Ni %A Xin Li %A Michael Qizhe Shieh %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-chen25s %I PMLR %P 8093--8107 %U https://proceedings.mlr.press/v267/chen25s.html %V 267 %X The emergence of long-context large language models (LLMs) offers a promising alternative to traditional retrieval-augmented generation (RAG) for processing extensive documents. However, the computational overhead of long-context inference presents significant efficiency challenges. While Speculative Decoding (SD) traditionally accelerates inference using smaller draft models, its effectiveness diminishes substantially in long-context scenarios due to memory-bound KV cache operations. We introduce Retrieval-Augmented Speculative Decoding (RAPID), which leverages RAG for both accelerating and enhancing generation quality in long-context inference. RAPID introduces the RAG drafter—a draft LLM operating on shortened retrieval contexts—to speculate on the generation of long-context target LLMs. Our approach enables a new paradigm where same-scale or even larger LLMs can serve as RAG drafters while maintaining computational efficiency. To fully leverage the potentially superior capabilities from stronger RAG drafters, we develop an inference-time knowledge transfer that enriches the target distribution by RAG. Extensive experiments on the LLaMA-3.1 and Qwen2.5 backbones demonstrate that RAPID effectively integrates the strengths of both RAG and long-context LLMs, achieving significant performance improvements (e.g., from 39.33 to 42.83 on InfiniteBench for LLaMA-3.1-8B) with more than 2$\times$ speedups for long-context inference. Our analyses also reveal the robustness of RAPID across various context lengths and retrieval quality.
APA
Chen, G., Feng, Q., Ni, J., Li, X. & Shieh, M.Q.. (2025). RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:8093-8107 Available from https://proceedings.mlr.press/v267/chen25s.html.

Related Material