Loïc Cabannes

Gergely Szilvasy, Manuel Faysse, Maria Lomeli et al.

Under modern test-time compute and agentic paradigms, language models process ever-longer sequences. Efficient text generation with transformer architectures is increasingly constrained by the Key-Value cache memory footprint and bandwidth. To address this limitation, we introduce Self-Pruned Key-Value Attention (SP-KV), a mechanism designed to predict future KV utility in order to reduce the size of the long-term KV cache. This strategy operates at a fine granularity: a lightweight utility predictor scores each key-value pair, and while recent KVs are always available via a local window, older pairs are written in the cache and used in global attention only if their predicted utility surpasses a given threshold. The LLM and the utility predictor are trained jointly end-to-end exclusively through next-token prediction loss, and are adapted from pretrained LLM checkpoints. Rather than enforcing a fixed compression ratio, SP-KV performs dynamic sparsification: the mechanism adapts to the input and typically reduces the KV cache size by a factor of $3$ to $10\times$, longer sequences often being more compressible. This leads to vast improvements in memory usage and decoding speed, with little to no degradation of validation loss nor performance on a broad set of downstream tasks. Beyond serving as an effective KV-cache reduction mechanism, our method reveals structured layer- and head-specific sparsity patterns that we can use to guide the design of hybrid local-global attention architectures.

Loïc Cabannes, Maximilian Beck, Gergely Szilvasy et al.

Recent works show that hybrid architectures combining sliding window softmax attention layers with linear recurrent neural network (RNN) layers outperform both of these architectures taken separately. However, the impact of the window length and the interplay between softmax attention and linear RNN layers remain under-studied. In this work, we introduce SWAX, a hybrid architecture consisting of sliding-window attention and xLSTM linear RNN layers. A counter-intuitive finding with SWAX is that larger sliding windows do not improve the long-context performance. In fact, short window attention encourages the model to better train the long-term memory of the xLSTM, by relying less on the softmax attention mechanism for long context-retrieval. The issue with small sliding windows is that they are detrimental for short-context tasks, which could be solved with information from moderately larger sliding windows otherwise. Therefore, we train SWAX by stochastically changing the sliding window size, forcing the model to leverage both a longer context window and the xLSTM memory. SWAX trained with stochastic window sizes significantly outperforms regular window attention both on short and long-context problems.

Swarnadeep Bhar, Omar Naim, Eleni Metheniti et al.

We present COCORELI, a hybrid agent framework designed to tackle the limitations of large language models (LLMs) in tasks requiring: following complex instructions, minimizing hallucination, and spatial reasoning. COCORELI integrates medium-sized LLM agents with novel abstraction mechanisms and a discourse module to parse instructions to in-context learn dynamic, high-level representations of the environment. Experiments on natural collaborative construction tasks show that COCORELI outperforms single-LLM CoT and agentic LLM systems, all using larger LLMs. It manages to largely avoid hallucinations, identify missing information, ask for clarifications, and update its learned objects. COCORELI's abstraction abilities extend beyond ENVIRONMENT, as shown in the ToolBench API completion task.