SparQ Attention: Bandwidth-Efficient LLM Inference

Luka Ribar, Ivan Chelombiev, Luke Hudlass-Galley, Charlie Blake, Carlo Luschi, Douglas Orr
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:42558-42583, 2024.

Abstract

The computational difficulties of large language model (LLM) inference remain a significant obstacle to their widespread deployment. The need for many applications to support long input sequences and process them in large batches typically causes token-generation to be bottlenecked by data transfer. For this reason, we introduce SparQ Attention, a technique for increasing the inference throughput of LLMs by utilising memory bandwidth more efficiently within the attention layers, through selective fetching of the cached history. Our proposed technique can be applied directly to off-the-shelf LLMs during inference, without requiring any modification to the pre-training setup or additional fine-tuning. We show that SparQ Attention brings up to 8x savings in attention data transfers without substantial drops in accuracy, by evaluating Llama 2 and 3, Mistral, Gemma and Pythia models on a wide range of downstream tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-ribar24a, title = {{S}par{Q} Attention: Bandwidth-Efficient {LLM} Inference}, author = {Ribar, Luka and Chelombiev, Ivan and Hudlass-Galley, Luke and Blake, Charlie and Luschi, Carlo and Orr, Douglas}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {42558--42583}, 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/ribar24a/ribar24a.pdf}, url = {https://proceedings.mlr.press/v235/ribar24a.html}, abstract = {The computational difficulties of large language model (LLM) inference remain a significant obstacle to their widespread deployment. The need for many applications to support long input sequences and process them in large batches typically causes token-generation to be bottlenecked by data transfer. For this reason, we introduce SparQ Attention, a technique for increasing the inference throughput of LLMs by utilising memory bandwidth more efficiently within the attention layers, through selective fetching of the cached history. Our proposed technique can be applied directly to off-the-shelf LLMs during inference, without requiring any modification to the pre-training setup or additional fine-tuning. We show that SparQ Attention brings up to 8x savings in attention data transfers without substantial drops in accuracy, by evaluating Llama 2 and 3, Mistral, Gemma and Pythia models on a wide range of downstream tasks.} }
Endnote
%0 Conference Paper %T SparQ Attention: Bandwidth-Efficient LLM Inference %A Luka Ribar %A Ivan Chelombiev %A Luke Hudlass-Galley %A Charlie Blake %A Carlo Luschi %A Douglas Orr %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-ribar24a %I PMLR %P 42558--42583 %U https://proceedings.mlr.press/v235/ribar24a.html %V 235 %X The computational difficulties of large language model (LLM) inference remain a significant obstacle to their widespread deployment. The need for many applications to support long input sequences and process them in large batches typically causes token-generation to be bottlenecked by data transfer. For this reason, we introduce SparQ Attention, a technique for increasing the inference throughput of LLMs by utilising memory bandwidth more efficiently within the attention layers, through selective fetching of the cached history. Our proposed technique can be applied directly to off-the-shelf LLMs during inference, without requiring any modification to the pre-training setup or additional fine-tuning. We show that SparQ Attention brings up to 8x savings in attention data transfers without substantial drops in accuracy, by evaluating Llama 2 and 3, Mistral, Gemma and Pythia models on a wide range of downstream tasks.
APA
Ribar, L., Chelombiev, I., Hudlass-Galley, L., Blake, C., Luschi, C. & Orr, D.. (2024). SparQ Attention: Bandwidth-Efficient LLM Inference. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:42558-42583 Available from https://proceedings.mlr.press/v235/ribar24a.html.

Related Material