With the continuous advancement in the performance of large language models (LLMs), their demand for computational resources and memory has significantly increased, which poses major challenges for efficient inference on consumer-grade devices and legacy servers. These devices typically feature relatively weaker GPUs and stronger CPUs. Although techniques such as parameter offloading and partial offloading can alleviate GPU memory pressure to some extent, their effectiveness is limited due to communication latency and suboptimal hardware resource utilization. To address this issue, we propose Dovetail—a lossless inference acceleration method that leverages the complementary characteristics of heterogeneous devices and the advantages of speculative decoding. Dovetail deploys a draft model on the GPU to perform preliminary predictions, while a target model running on the CPU validates these outputs. By reducing the granularity of data transfer, Dovetail significantly minimizes communication overhead. To further improve efficiency, we optimize the draft model specifically for heterogeneous hardware environments by reducing the number of draft tokens to lower parallel verification latency, increasing model depth to enhance predictive capabilities, and introducing a Dynamic Gating Fusion (DGF) mechanism to improve the integration of feature and embedding information. We conduct comprehensive evaluations of Dovetail across various consumer-grade GPUs, covering multiple tasks and mainstream models. Experimental results on 13B models demonstrate that Dovetail achieves inference speedups ranging from 1.79× to 10.1× across different devices, while maintaining consistency and stability in the distribution of generated texts.

Outline generation aims to uncover the internal content structure of a document by identifying potential chapter connections and generating corresponding summaries. A robust outline generation model strives for coherence between and within plots. However, existing methods perform well on short- and medium-length texts and struggle with generating readable outlines for very long texts (e.g., fictional literary works). The primary challenge lies in their inability to accurately segment plots within long texts. To address this issue, we propose a novel unsupervised guidance framework, LeStrTP, to guide large language model (LLM) outline generation. This framework ensures that each structured plot encapsulates complete causality by accurately identifying plot boundaries. Specifically, the LeStrTP framework constructs chapter-level graph from long texts and learns their embeddings. Subsequently, through Markov chain modeling chapter dependence, a unique search operator is designed to achieve plot segmentation. To facilitate research on this task, we introduce a new annotated benchmark dataset, NovOutlineSet. Experimental results demonstrate that structured plots not only enhance the coherence and integrity of generated outlines but also significantly improve their quality.