Deliberation in Latent Space via Differentiable Cache Augmentation

Luyang Liu, Jonas Pfeiffer, Jiaxing Wu, Jun Xie, Arthur Szlam
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:39261-39274, 2025.

Abstract

Techniques enabling large language models (LLMs) to "think more" by generating and attending to intermediate reasoning steps have shown promise in solving complex problems. However, the standard approaches generate sequences of discrete tokens immediately before responding, and so they can incur significant latency costs and be challenging to optimize. In this work, we demonstrate that a frozen LLM can be augmented with an offline coprocessor that operates on the model’s key-value (kv) cache. This coprocessor augments the cache with a set of latent embeddings designed to improve the fidelity of subsequent decoding. We train this coprocessor using the language modeling loss from the decoder on standard pretraining data, while keeping the decoder itself frozen. This approach enables the model to learn, in an end-to-end differentiable fashion, how to distill additional computation into its kv-cache. Because the decoder remains unchanged, the coprocessor can operate offline and asynchronously, and the language model can function normally if the coprocessor is unavailable or if a given cache is deemed not to require extra computation. We show experimentally that when a cache is augmented, the decoder achieves lower perplexity on numerous subsequent tokens. Furthermore, even without any task-specific training, our experiments demonstrate that cache augmentation consistently improves performance across a range of reasoning-intensive tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-liu25bc, title = {Deliberation in Latent Space via Differentiable Cache Augmentation}, author = {Liu, Luyang and Pfeiffer, Jonas and Wu, Jiaxing and Xie, Jun and Szlam, Arthur}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {39261--39274}, 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/liu25bc/liu25bc.pdf}, url = {https://proceedings.mlr.press/v267/liu25bc.html}, abstract = {Techniques enabling large language models (LLMs) to "think more" by generating and attending to intermediate reasoning steps have shown promise in solving complex problems. However, the standard approaches generate sequences of discrete tokens immediately before responding, and so they can incur significant latency costs and be challenging to optimize. In this work, we demonstrate that a frozen LLM can be augmented with an offline coprocessor that operates on the model’s key-value (kv) cache. This coprocessor augments the cache with a set of latent embeddings designed to improve the fidelity of subsequent decoding. We train this coprocessor using the language modeling loss from the decoder on standard pretraining data, while keeping the decoder itself frozen. This approach enables the model to learn, in an end-to-end differentiable fashion, how to distill additional computation into its kv-cache. Because the decoder remains unchanged, the coprocessor can operate offline and asynchronously, and the language model can function normally if the coprocessor is unavailable or if a given cache is deemed not to require extra computation. We show experimentally that when a cache is augmented, the decoder achieves lower perplexity on numerous subsequent tokens. Furthermore, even without any task-specific training, our experiments demonstrate that cache augmentation consistently improves performance across a range of reasoning-intensive tasks.} }
Endnote
%0 Conference Paper %T Deliberation in Latent Space via Differentiable Cache Augmentation %A Luyang Liu %A Jonas Pfeiffer %A Jiaxing Wu %A Jun Xie %A Arthur Szlam %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-liu25bc %I PMLR %P 39261--39274 %U https://proceedings.mlr.press/v267/liu25bc.html %V 267 %X Techniques enabling large language models (LLMs) to "think more" by generating and attending to intermediate reasoning steps have shown promise in solving complex problems. However, the standard approaches generate sequences of discrete tokens immediately before responding, and so they can incur significant latency costs and be challenging to optimize. In this work, we demonstrate that a frozen LLM can be augmented with an offline coprocessor that operates on the model’s key-value (kv) cache. This coprocessor augments the cache with a set of latent embeddings designed to improve the fidelity of subsequent decoding. We train this coprocessor using the language modeling loss from the decoder on standard pretraining data, while keeping the decoder itself frozen. This approach enables the model to learn, in an end-to-end differentiable fashion, how to distill additional computation into its kv-cache. Because the decoder remains unchanged, the coprocessor can operate offline and asynchronously, and the language model can function normally if the coprocessor is unavailable or if a given cache is deemed not to require extra computation. We show experimentally that when a cache is augmented, the decoder achieves lower perplexity on numerous subsequent tokens. Furthermore, even without any task-specific training, our experiments demonstrate that cache augmentation consistently improves performance across a range of reasoning-intensive tasks.
APA
Liu, L., Pfeiffer, J., Wu, J., Xie, J. & Szlam, A.. (2025). Deliberation in Latent Space via Differentiable Cache Augmentation. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:39261-39274 Available from https://proceedings.mlr.press/v267/liu25bc.html.

Related Material