SCOPE is a training‑free generative prompt‑compression framework that reduces LLM input length by chunking a prompt into semantically coherent segments, rewriting each chunk to be more concise, and then reconstructing a coherent prompt. Unlike token‑removal methods, SCOPE’s chunk‑level rewriting preserves critical information and text coherence, and includes optimization techniques for finer‑grained control of compression ratios. Extensive evaluations on question‑answering and summarization tasks show that SCOPE consistently outperforms selective compression baselines, especially at high compression ratios.
arXiv:2511.23271v2 Announce Type: replace Abstract: Long system prompts are widely used to steer Large Language Models (LLMs), but repeatedly processing them at inference time is inefficient and cons...
The paper introduces SCSP, a training‑free framework that improves long‑context embeddings by selectively pooling informative tokens. SCSP partitions documents into sentence‑aware chunks, adds a semantic compression prompt to each chunk, and uses prompt‑isolated attention masks to estimate token importance. The selected tokens’ intermediate‑layer representations are aggregated to form the final embedding, yielding consistent performance gains across zero‑shot and fine‑tuned models on long‑context benchmarks.
The paper investigates prompt minimization, aiming to reduce prompts to their smallest, most information-dense form without losing output fidelity. It argues that shorter prompts lower computational overhead and inference latency, especially when large contexts are unnecessarily included, and that longer prompts can harm LLM reasoning and accuracy. The authors propose three frameworks to identify minimal prompts and show that these often produce outputs comparable to longer versions, highlighting redundancy in the input space and opening new avenues for efficient prompt engineering.
The paper introduces K-Token Merging, a latent-space compression method that merges each contiguous block of K token embeddings into a single embedding using a lightweight encoder. The compressed sequence is then processed by a LoRA-adapted large language model, while generation continues in the original vocabulary. Experiments on tasks such as structural reasoning, sentiment classification, and code editing demonstrate that K-Token Merging achieves up to 75% input length reduction with minimal performance loss, placing it on the Pareto frontier of performance versus compression.
Doc2FRC introduces Fixed-Range Chunking (FRC), a dynamic programming method that partitions documents into chunks of a predefined length interval, ensuring consistent length distributions during training and inference. This approach reduces train-test length mismatch, mitigates n-gram repetition, and improves translation quality for 7B LLMs compared to direct Doc2Doc fine-tuning. Experiments on IWSLT2017 and a new 10-language test set, GlobVDoc, demonstrate that FRC outperforms existing document-level machine translation methods and enhances out-of-distribution translation performance.
The paper presents a survey of 129 public large language model (LLM) prompt datasets, totaling over 1.22 TB and 673 million instances, and introduces a unified taxonomy for them. By analyzing seven datasets in depth, the authors identify lexical, syntactic, and semantic patterns that differentiate prompts from general text, and evaluate these patterns for tasks such as prompt filtering, source domain routing, and response quality assessment. They demonstrate that a 63‑dimensional linguistic feature set extracted on a CPU can match over 91 % of the F1 score of GPU‑based sentence embeddings while halving latency, and that structural features can effectively route prompts across datasets, though they may negatively impact response quality when prompt length is controlled.
arXiv:2609.13154v1 Announce Type: new Abstract: Recent advances in large language models (LLMs) have made prompts increasingly large and complex. Techniques such as chain-of-thought reasoning (Wei et...
arXiv:2607. 17486v1 Announce Type: cross Abstract: As large language models (LLMs) process increasingly longer prompts, computation and KV-cache memory costs have emerged as major bottlenecks in inference systems.
arXiv:2608. 13588v1 Announce Type: cross Abstract: Multi-hop question answering requires complex reasoning across multiple evidence segments, which often overwhelms retrieval-augmented generation systems with lengthy and noisy contexts, thereby undermining both efficiency and accuracy.
The paper introduces Highlight-Then-Summarize (H2S), a two-step approach that first highlights question-relevant evidence in long documents and then condenses it into a compact, question-conditioned summary before generating an answer. The authors built the H2S-Dataset with 6,647 examples spanning 11 benchmark families, and developed H2S-RL to reward evidence selection and summary construction. Evaluated on the H2S-Bench suite, the H2S-14B model outperforms larger open-source models, achieving the highest overall score and maintaining strong performance even with a reduced output budget.
The paper introduces CARPAS, a new task that dynamically refines user-provided aspects for aspect-based summarization in large language models (LLMs). It presents three new datasets and evaluates four prompting strategies, finding that LLMs tend to over-generate aspects, leading to overly long and misaligned summaries. To address this, the authors propose a two-stage framework that first generates lightweight scope guidance before aspect refinement and summarization, which improves focus, reduces over-generation, and enhances performance across all datasets.
CoEM introduces a Commit-on-Evidence Memory system that learns when to compress source evidence into compact memory facts while preserving potentially useful excerpts verbatim in a pending set. The system uses a learned policy to decide whether to promote, retain, or discard each pending excerpt as new context arrives, and a frozen verifier ensures only supported facts are committed. Reinforcement learning trains this policy with step-level evidence rewards and final answer rewards, leading to consistent improvements in long-context reasoning, achieving 10.4–11.4 F1 points over the strongest baseline on 6,400-document inputs.