Rapid advances in Large Language Models (LLMs) have spurred demand for processing extended context sequences in contemporary applications. However, this progress faces two challenges: performance degradation due to sequence lengths out-of-distribution, and excessively long inference times caused by the quadratic computational complexity of attention. These issues limit LLMs in long-context scenarios. In this paper, we propose Dynamic Token-Level KV Cache Selection (*TokenSelect*), a training-free method for efficient and accurate long-context inference. *TokenSelect* builds upon the observation of non-contiguous attention sparsity, using QK dot products to measure per-head KV Cache criticality at token-level. By per-head soft voting mechanism, *TokenSelect* selectively involves a few critical KV cache tokens in attention calculation without sacrificing accuracy. To further accelerate *TokenSelect*, we design the Selection Cache based on observations of consecutive Query similarity and implemented the efficient Paged Dot Product Kernel, significantly reducing the selection overhead. A comprehensive evaluation of *TokenSelect* demonstrates up to 23.84× speedup in attention computation and up to 2.28× acceleration in end-to-end latency, while providing superior performance compared to state-of-the-art long-context inference methods.

Generative query rewrite generates reconstructed query rewrites using the conversation history while rely heavily on gold rewrite pairs that are expensive to obtain. Recently, few-shot learning is gaining increasing popularity for this task, whereas these methods are sensitive to the inherent noise due to limited data size. Besides, both attempts face performance degradation when there exists language style shift between training and testing cases. To this end, we study low-resource generative conversational query rewrite that is robust to both noise and language style shift. The core idea is to utilize massive unlabeled data to make further improvements via a contrastive co-training paradigm. Specifically, we co-train two dual models (namely Rewriter and Simplifier) such that each of them provides extra guidance through pseudo-labeling for enhancing the other in an iterative manner. We also leverage contrastive learning with data augmentation, which enables our model pay more attention on the truly valuable information than the noise. Extensive experiments demonstrate the superiority of our model under both few-shot and zero-shot scenarios. We also verify the better generalization ability of our model when encountering language style shift.

The task of query rewrite aims to convert an in-context query to its fully-specified version where ellipsis and coreference are completed and referred-back according to the history context. Although much progress has been made, less efforts have been paid to real scenario conversations that involve drawing information from more than one modalities. In this paper, we propose the task of multimodal conversational query rewrite (McQR), which performs query rewrite under the multimodal visual conversation setting. We collect a large-scale dataset named McQueen based on manual annotation, which contains 15k visual conversations and over 80k queries where each one is associated with a fully-specified rewrite version. In addition, for entities appearing in the rewrite, we provide the corresponding image box annotation. We then use the McQueen dataset to benchmark a state-of-the-art method for effectively tackling the McQR task, which is based on a multimodal pre-trained model with pointer generator. Extensive experiments are performed to demonstrate the effectiveness of our model on this task.