AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism

Zhepei Wei, Wei-Lin Chen, Xinyu Zhu, Yu Meng
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:65981-65996, 2025.

Abstract

Large language models (LLMs) are increasingly used for long-content generation (e.g., long Chain-of-Thought reasoning) where decoding efficiency becomes a critical bottleneck: Autoregressive decoding is inherently limited by its sequential token generation process, where each token must be generated before the next can be processed. This sequential dependency restricts the ability to fully leverage modern hardware’s parallel processing capabilities. Existing methods like speculative decoding and layer skipping offer potential speedups but have notable drawbacks: speculative decoding relies on an auxiliary “drafter” model, which can be challenging to acquire and increases memory overhead, while layer skipping may introduce discrepancies in the generated outputs due to the missing key-value cache at skipped layers. In this work, we propose AdaDecode, which accelerates LLM decoding without requiring auxiliary models or changes to the original model parameters, while ensuring output consistency. AdaDecode leverages the insight that many tokens—particularly simple or highly-predictable ones—can accurately be generated at intermediate layers, as further layers often do not significantly alter predictions once the model reaches a certain confidence. By adaptively generating tokens at intermediate layers when confidence is high, AdaDecode enables the next token’s computation to begin immediately. The remaining layer computations for early-predicted tokens are deferred and executed in parallel with subsequent tokens when needed, maximizing hardware utilization and reducing decoding latency. A final verification step ensures that early predictions match the results of standard autoregressive decoding, preserving output parity. Experiments across diverse generation tasks shows that AdaDecode consistently achieves superior decoding throughput compared to baselines with up to 1.73$\times$ speedup, while guaranteeing output parity with standard autoregressive decoding.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-wei25a, title = {{A}da{D}ecode: Accelerating {LLM} Decoding with Adaptive Layer Parallelism}, author = {Wei, Zhepei and Chen, Wei-Lin and Zhu, Xinyu and Meng, Yu}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {65981--65996}, 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/wei25a/wei25a.pdf}, url = {https://proceedings.mlr.press/v267/wei25a.html}, abstract = {Large language models (LLMs) are increasingly used for long-content generation (e.g., long Chain-of-Thought reasoning) where decoding efficiency becomes a critical bottleneck: Autoregressive decoding is inherently limited by its sequential token generation process, where each token must be generated before the next can be processed. This sequential dependency restricts the ability to fully leverage modern hardware’s parallel processing capabilities. Existing methods like speculative decoding and layer skipping offer potential speedups but have notable drawbacks: speculative decoding relies on an auxiliary “drafter” model, which can be challenging to acquire and increases memory overhead, while layer skipping may introduce discrepancies in the generated outputs due to the missing key-value cache at skipped layers. In this work, we propose AdaDecode, which accelerates LLM decoding without requiring auxiliary models or changes to the original model parameters, while ensuring output consistency. AdaDecode leverages the insight that many tokens—particularly simple or highly-predictable ones—can accurately be generated at intermediate layers, as further layers often do not significantly alter predictions once the model reaches a certain confidence. By adaptively generating tokens at intermediate layers when confidence is high, AdaDecode enables the next token’s computation to begin immediately. The remaining layer computations for early-predicted tokens are deferred and executed in parallel with subsequent tokens when needed, maximizing hardware utilization and reducing decoding latency. A final verification step ensures that early predictions match the results of standard autoregressive decoding, preserving output parity. Experiments across diverse generation tasks shows that AdaDecode consistently achieves superior decoding throughput compared to baselines with up to 1.73$\times$ speedup, while guaranteeing output parity with standard autoregressive decoding.} }
Endnote
%0 Conference Paper %T AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism %A Zhepei Wei %A Wei-Lin Chen %A Xinyu Zhu %A Yu Meng %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-wei25a %I PMLR %P 65981--65996 %U https://proceedings.mlr.press/v267/wei25a.html %V 267 %X Large language models (LLMs) are increasingly used for long-content generation (e.g., long Chain-of-Thought reasoning) where decoding efficiency becomes a critical bottleneck: Autoregressive decoding is inherently limited by its sequential token generation process, where each token must be generated before the next can be processed. This sequential dependency restricts the ability to fully leverage modern hardware’s parallel processing capabilities. Existing methods like speculative decoding and layer skipping offer potential speedups but have notable drawbacks: speculative decoding relies on an auxiliary “drafter” model, which can be challenging to acquire and increases memory overhead, while layer skipping may introduce discrepancies in the generated outputs due to the missing key-value cache at skipped layers. In this work, we propose AdaDecode, which accelerates LLM decoding without requiring auxiliary models or changes to the original model parameters, while ensuring output consistency. AdaDecode leverages the insight that many tokens—particularly simple or highly-predictable ones—can accurately be generated at intermediate layers, as further layers often do not significantly alter predictions once the model reaches a certain confidence. By adaptively generating tokens at intermediate layers when confidence is high, AdaDecode enables the next token’s computation to begin immediately. The remaining layer computations for early-predicted tokens are deferred and executed in parallel with subsequent tokens when needed, maximizing hardware utilization and reducing decoding latency. A final verification step ensures that early predictions match the results of standard autoregressive decoding, preserving output parity. Experiments across diverse generation tasks shows that AdaDecode consistently achieves superior decoding throughput compared to baselines with up to 1.73$\times$ speedup, while guaranteeing output parity with standard autoregressive decoding.
APA
Wei, Z., Chen, W., Zhu, X. & Meng, Y.. (2025). AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:65981-65996 Available from https://proceedings.mlr.press/v267/wei25a.html.

Related Material