Searcharxiv⌕ Search

arXiv subjects

Yiqi Li

Publications and source records attributed to Yiqi Li.

8 recordsLinked to original sources
arXiv2026

Structural Process Supervision for Latent Chain-of-Thought Reasoning

Latent reasoning approaches enhance token-level efficiency and robustness by replacing verbose, explicit chain-of-thought (CoT) tokens with compact continuous-space embeddings. However, existing methods lack direct process supervision over these latent embeddings, which often leads to representation collapse and uneven information distribution. To address this, we propose Prototype-Mediated Process Supervision (PMPS), which introduces learnable reasoning prototypes as semantic anchors to provide structural process-level supervision for latent reasoning. PMPS projects latent embeddings and explicit CoT embeddings into a shared prototype space, achieving many-to-many soft alignment between unequal-length representations through prototype assignment. Meanwhile, we introduce a Progressive Sequential Alignment (PSA) module to further guide training: positional priors initially encourage sequential alignment structure, then gradually relax to permit adaptive matching. Experimental results show that PMPS compresses output token length to under 50% of explicit CoT on GSM8K-Aug. Compared to leading baseline SIM-CoT, our method achieves average accuracy gains of 2.08% across different model families. On GPT-2, PMPS even surpasses CoT-SFT. On larger models and a more challenging task, PMPS consistently attains the highest accuracy among all latent reasoning methods with comparable output length.

cs.AI↗
arXiv2026

PRISM: Fast Online LLM Serving via Scheduling-Memory Co-design

Modern online large language model (LLM) services, such as Retrieval-Augmented Generation (RAG) and agent systems, increasingly expose two prominent characteristics: prompt segmentation (e.g., system instructions, retrieved passages, tool outputs) and hotspot skew, where a small set of these segments recurs frequently across user requests. Failing to jointly exploit these patterns could lead to repeated prefill of hot segments and prolonged TTFT, undermining both throughput and user-perceived responsiveness. However, existing work tackles these patterns independently: KV-cache management mainly exploits segment reuse while scheduling reorders requests to improve cache locality, yet neither aligns request admission with KV-cache retention. To address this gap, we first analyze how scheduling and KV-cache management jointly affect TTFT. Guided by this, we present PRISM (Prefix Reuse Optimization Integrated Scheduling and Memory), which co-designs a query-aware scheduler (QAS) with a demand-aware radix tree (DART) to align request admission with exact-prefix KV retention. Our evaluation results show that, versus the strongest baseline, PRISM reduces average per-QPS P99 TTFT by 23.3\% and 37.1\% while increasing exact-prefix KV-cache hit rate by 5.9 and 12.2 percentage points on 4B and 13B models, respectively.

cs.LG↗
arXiv2025

VocalBench-zh: Decomposing and Benchmarking the Speech Conversational Abilities in Mandarin Context

The development of multi-modal large language models (LLMs) leads to intelligent approaches capable of speech interactions. As one of the most widely spoken languages globally, Mandarin is supported by most models to enhance their applicability and reach. However, the scarcity of comprehensive speech-to-speech (S2S) benchmarks in Mandarin contexts impedes systematic evaluation for developers and hinders fair model comparison for users. In this work, we propose VocalBench-zh, an ability-level divided evaluation suite adapted to Mandarin context consisting of 10 well-crafted subsets and over 10K high-quality instances, covering 12 user-oriented characters. The evaluation experiment on 14 mainstream models reveals the common challenges for current routes, and highlights the need for new insights into next-generation speech interactive systems. The evaluation codes and datasets will be available at https://github.com/SJTU-OmniAgent/VocalBench-zh.

cs.CL↗
arXiv2025

DICE: Structured Reasoning in LLMs through SLM-Guided Chain-of-Thought Correction

When performing reasoning tasks with user-specific requirements, such as strict output formats, large language models (LLMs) often prioritize reasoning over adherence to detailed instructions. Fine-tuning LLMs on supervised datasets to address this is impractical due to high computational costs and limited parameter access. To tackle this, we propose DICE, a lightweight framework that guides small language models (SLMs) to refine LLMs' outputs through chain-of-thought (CoT) correction. DICE decouples the process by first prompting LLMs to generate natural language responses, then using trained SLMs to analyze and refine these outputs to meet structured output specifications. This framework preserves LLMs' broad knowledge and reasoning capabilities while ensuring the outputs conform to user demands. Specifically, DICE first constructs structured CoT adaptation datasets via a two-stage method and subsequently applies a dual-tuning strategy to fine-tune SLMs for generating structured outputs in an analyze-then-answer pattern. Experiments demonstrate that DICE improves the average format accuracy and content correctness of LLM outputs by 35.4\% and 29.4\%, respectively, achieving state-of-the-art (SOTA) performance over other competitive baselines.

cs.CL↗
arXiv2025

How Growing Toxicity Manifests: A Topic Trajectory Analysis of U.S. Immigration Discourse on Social Media

In the online public sphere, discussions about immigration often become increasingly fractious, marked by toxic language and polarization. Drawing on 4 million X posts over six months, we combine a user- and topic-centric approach to study how shifts in toxicity manifest as topical shifts. Our topic discovery method, which leverages instruction-based embeddings and recursive HDBSCAN, uncovers 157 fine-grained subtopics within the U.S. immigration discourse. We focus on users in four groups: (1) those with increasing toxicity, (2) those with decreasing toxicity, and two reference groups with no significant toxicity trend but matched toxicity levels. Treating each posting history as a trajectory through a five-dimensional topic space, we compare average group trajectories using permutational MANOVA. Our findings show that users with increasing toxicity drift toward alarmist, fear-based frames, whereas those with decreasing toxicity pivot toward legal and policy-focused themes. Both patterns diverge statistically significantly from their reference groups. This pipeline, which combines hierarchical topic discovery with trajectory analysis, offers a replicable method for studying dynamic conversations around social issues at scale.

cs.SI↗
arXiv2025

VocalBench: Benchmarking the Vocal Conversational Abilities for Speech Interaction Models

Speech large language models (SpeechLLMs) have extended human-machine interactions from the text modality to the dynamic speech domain. Spoken dialogues convey diverse information, including semantic concepts, acoustic variations, paralanguage cues, and environmental context. However, existing evaluations of speech interaction models lack instances mimicking real scenarios and predominantly focus on the performance of distinct aspects, lacking a comprehensive comparison of critical capabilities between current routines. To address this gap, we propose VocalBench to assess the speech conversational abilities, comprising around 24k carefully curated instances of both English and Mandarin across four key dimensions - semantic quality, acoustic performance, conversational abilities, and robustness, covering 14 user-oriented characters. Experiments on 27 mainstream models reveal the common challenges for current routes, and highlight the need for new insights into next-generation speech interactive systems.

cs.CL↗
arXiv2024

Approximate Vector Set Search Inspired by Fly Olfactory Neural System

Vector set search, an underexplored similarity search paradigm, aims to find vector sets similar to a query set. This search paradigm leverages the inherent structural alignment between sets and real-world entities to model more fine-grained and consistent relationships for diverse applications. This task, however, faces more severe efficiency challenges than traditional single-vector search due to the combinatorial explosion of pairings in set-to-set comparisons. In this work, we aim to address the efficiency challenges posed by the combinatorial explosion in vector set search, as well as the curse of dimensionality inherited from single-vector search. To tackle these challenges, we present an efficient algorithm for vector set search, BioVSS (Bio-inspired Vector Set Search). BioVSS simulates the fly olfactory circuit to quantize vectors into sparse binary codes and then designs an index based on the set membership property of the Bloom filter. The quantization and indexing strategy enables BioVSS to efficiently perform vector set search by pruning the search space. Experimental results demonstrate over 50 times speedup compared to linear scanning on million-scale datasets while maintaining a high recall rate of up to 98.9%, making it an efficient solution for vector set search.

cs.DB↗