Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads

Tianle Cai, Yuhong Li, Zhengyang Geng, Hongwu Peng, Jason D. Lee, Deming Chen, Tri Dao
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:5209-5235, 2024.

Abstract

Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one’s output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator’s cache. While methods such as speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the model’s capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2$\times$ speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-2.8$\times$.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-cai24b, title = {Medusa: Simple {LLM} Inference Acceleration Framework with Multiple Decoding Heads}, author = {Cai, Tianle and Li, Yuhong and Geng, Zhengyang and Peng, Hongwu and Lee, Jason D. and Chen, Deming and Dao, Tri}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {5209--5235}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/cai24b/cai24b.pdf}, url = {https://proceedings.mlr.press/v235/cai24b.html}, abstract = {Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one’s output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator’s cache. While methods such as speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the model’s capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2$\times$ speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-2.8$\times$.} }
Endnote
%0 Conference Paper %T Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads %A Tianle Cai %A Yuhong Li %A Zhengyang Geng %A Hongwu Peng %A Jason D. Lee %A Deming Chen %A Tri Dao %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-cai24b %I PMLR %P 5209--5235 %U https://proceedings.mlr.press/v235/cai24b.html %V 235 %X Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one’s output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator’s cache. While methods such as speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the model’s capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2$\times$ speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-2.8$\times$.
APA
Cai, T., Li, Y., Geng, Z., Peng, H., Lee, J.D., Chen, D. & Dao, T.. (2024). Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:5209-5235 Available from https://proceedings.mlr.press/v235/cai24b.html.

Related Material