Context compression has emerged as a vital approach to mitigate both performance degradation and resource consumption for large language models (LLMs) in long context scenarios. Current state-of-the-art methods allocate the compression budget at fixed-size block granularity, either uniformly or with adaptation across blocks. Both strategies fragment sentence-level semantic units and overlook the relevance variation within each block. In this paper, we propose the Adaptive Semantic Preservation (ASP) compressor, built on two innovations: (1) To preserve semantic units, our Semantic-Preserving Segmentation partitions context along natural sentence boundaries while maintaining the morphological atomicity of words. (2) To prioritize critical information, our task-aware Adaptive Budget Allocation dynamically assigns compression budget based on each sentence’s informational salience for the task instruction. Extensive experiments across diverse long-context scenarios show that ASP outperforms strong task-aware and task-agnostic compression baselines on both Llama-3-8B-Instruct and Qwen-2.5-7B-Instruct. In particular, ASP delivers stronger robustness compared with other baselines even at the extreme 16× high compression ratio.

Personalized dialogue agents require continuous long-term memory to maintain coherent interactions across multiple sessions. However, deploying these capabilities on consumer-grade hardware (e.g., 8 GB VRAM edge devices) introduces severe memory and compute bottlenecks. Existing systems typically rely on isotropic cosine similarity for retrieval and heuristic rules for context compression. These approaches lack a unified theoretical foundation, frequently suffering from the hubness problem in high-dimensional retrieval and syntactic fragmentation during compression. To overcome these limitations, we propose CoreMem, a resource-efficient edge-cloud memory architecture fundamentally unified by information geometry. First, Riemannian retrieval replaces cosine matching with a locally adaptive Fisher-Rao metric, effectively penalizing hub memories via Mahalanobis distance with O(Ndr) Woodbury acceleration for real-time search. Second, Fisher-guided discrete token distillation (FDTD) introduces a hierarchical sentence-to-token compression mechanism. It derives sensitivity scores from Fisher information traces, providing a principled compression-KL tradeoff augmented with explicit structural syntax protection. Evaluated on the LOCOMO and LongMemEval-S benchmarks, CoreMem achieves strong accuracy improvements, yielding substantial gains in Open-domain (+4.51 pp) and Temporal (+4.17 pp) reasoning. Extensive profiling confirms that the lightweight MiniLM-L6 configuration of CoreMem operates within a strict 8 GB VRAM budget, successfully bridging the gap between resource-constrained edge devices and the demand for theoretically grounded, lifelong memory agents.

Large Language Models (LLMs) face computational inefficiencies and redundant processing when handling long context inputs, prompting a focus on compression techniques. While existing semantic vector-based compression methods achieve promising performance, these methods fail to account for the intrinsic information density variations between context chunks, instead allocating soft tokens uniformly across context chunks. This uniform distribution inevitably diminishes allocation to information-critical regions. To address this, we propose Dynamic Allocation of Soft Tokens (DAST), a simple yet effective method that leverages the LLM’s intrinsic understanding of contextual relevance to guide compression. DAST combines perplexity-based local information with attention-driven global information to dynamically allocate soft tokens to the informative-rich chunks, enabling effective, context-aware compression. Experimental results across multiple benchmarks demonstrate that DAST surpasses state-of-the-art methods.

Writing assistance aims to improve the correctness and quality of input texts, with character checking being crucial in detecting and correcting wrong characters. In the real world where handwriting occupies the vast majority, characters that humans get wrong include faked characters (i.e., untrue characters created due to writing errors) and misspelled characters (i.e., true characters used incorrectly due to spelling errors). However, existing datasets and related studies only focus on misspelled characters that can be represented by computer text encoding systems, thereby ignoring faked characters which are more common and difficult. To break through this dilemma, we present Visual-C³, a human-annotated Visual Chinese Character Checking dataset with faked and misspelled Chinese characters. To the best of our knowledge, Visual-C³ is the first real-world visual and the largest human-crafted dataset for the Chinese character checking scenario. Additionally, we also propose and evaluate novel baseline methods on Visual-C³. Extensive empirical results and analyses show that Visual-C³ is high-quality yet challenging. As the first study focusing on Chinese faked characters, the dataset and the baseline methods are publicly available at https://github.com/THUKElab/Visual-C3.