ViCoStream: Streaming VideoLLMs Can Run Beyond 100 FPS with Stage-Wise Coordinated Inference

Streaming VideoLLMs must continuously process incoming video while maintaining low query latency, making both video-ingestion throughput and query-time responsiveness critical for real-time deployment. Existing methods largely focus on accelerating individual modules, such as visual encoding, token pruning, or KV-cache compression, but provide limited insight into whether the resulting system can sustain real-time streaming performance. We formulate streaming VideoLLM inference as a coordinated pipeline spanning visual preprocessing, visual encoding, token dropping, and LLM prefilling/decoding. Building on this formulation, we propose ViCoStream (Video Coordinated Streaming), a stage-wise coordinated streaming framework that combines chunk-wise execution, CUDA-stream overlap, visual token control, bounded visual attention, and query-side retrieval to bound per-chunk computation and memory costs. We further provide a systematic study of bottleneck migration, revealing how chunk size, token retention, attention locality, and retrieval scope shape the throughput-accuracy trade-off. Experiments with Qwen2.5-VL-3B/7B-Instruct across multiple streaming benchmarks show that ViCoStream achieves 134 FPS video throughput and less than 50 ms TTFT on a single A100 GPU while maintaining accuracy close to full-history baselines.

What Makes Effective Supervision in Latent Chain-of-Thought: An Information-Theoretic Analysis

Latent Chain-of-Thought (CoT) internalizes reasoning within continuous hidden states, offering a promising alternative to verbose discrete reasoning traces. However, robust latent reasoning remains difficult because outcome supervision provides weak learning signals and leaves latent trajectories prone to semantic drift. In this work, we analyze Latent CoT from an information-theoretic perspective and identify this failure as a dual collapse: gradient attenuation along the optimization path and representational drift in the latent space. We further decompose process supervision into two complementary dimensions: Trajectory Supervision, which injects dense stepwise reasoning signals, and Space Supervision, which preserves the semantic structure of the latent manifold. Our analysis shows that rigid geometric compression can collapse the reasoning space, whereas generative reconstruction provides a more flexible semantic anchor that better preserves information capacity. To measure these effects, we introduce the Unified Latent Probe (ULP), which quantifies the mutual information between latent trajectories and explicit reasoning steps. Experiments reveal a clear Information-Performance Binding: reasoning accuracy depends on the information fidelity preserved in the latent chain. These findings provide a principled framework for latent reasoning supervision and suggest shifting from geometric imitation toward mutual information maximization. Our code is available at \href{https://github.com/EIT-NLP/Supervision-in-Latent-CoT}{this repository}.

PowerOPD: Stabilizing On-Policy Distillation with Bounded Power Transformation

Standard on-policy distillation (OPD) for large language models estimates the reverse-KL objective using student-sampled tokens, yielding an unbiased single-sample Monte Carlo estimator that avoids vocabulary-wide computation. However, we show that this estimator suffers from severe training pathologies in practice: sample inefficiency, unstable generation dynamics, and a substantial performance gap compared to exact full-vocabulary OPD. Reward-level diagnosis traces these pathologies to the log-ratio reward, which is unbounded by construction, producing extremely high-variance gradients concentrated at early positions and persisting throughout training; standard post-hoc scaling fail as they operate only after this distortion occurs. To solve this problem, we propose PowerOPD: a family of natively bounded, sign-consistent rewards from the Box-Cox power transformation, parameterized by alpha > 0, of which the log-ratio is the degenerate alpha -> 0 limit. Across six mathematical reasoning benchmarks and four Qwen3 teacher-student pairs, PowerOPD achieves benchmark-averaged Avg@8/Pass@8 gains of up to +6.37/+5.71 over vanilla OPD, +3.01/+3.54 over post-hoc stabilization, and +2.59/+8.90 over full-vocabulary OPD, while reducing wall-clock time by 59.2% and peak GPU memory by 23.1%. Larger alpha generally improves accuracy, consistently shortens responses, and keeps gradient norms more than 3,000x smaller than vanilla OPD.

FlashNav: Ultra-Fast Policy Training for Robot Navigation within 20 Seconds

Deep reinforcement learning has shown strong potential for robot navigation, but its practical deployment is still limited by the long wall-clock cost of policy training. This paper presents FlashNav, a GPU-first framework for ultra-fast range-based robot navigation training. To the best of our knowledge, FlashNav is the first DRL-based robot navigation framework that reaches seconds-level policy training, with the fastest deployable policy trained in less than 20 seconds. The key idea is to align simulation with the navigation MDP: FlashNav preserves the essential components for velocity-level navigation, including occupancy geometry, range sensing, goal-conditioned control, robot motion dynamics, collision handling, termination, and reset, while removing unnecessary rendering and high-fidelity physical details from the training loop. Built on a batched bitmap simulator and a fully GPU-resident training pipeline with our FastDSAC learner, FlashNav generates massive parallel navigation transitions entirely on GPU. Experiments on TurtleBot2 and Unitree Go2 show that FlashNav achieves a 100\% success-rate below 20 seconds on an RTX 5090 and remains within tens of seconds across desktop GPUs. The learned policies further transfer to physical wheeled and legged robots in static and dynamic indoor scenes, demonstrating that DRL-based navigation can be trained at seconds-level speed while preserving deployable obstacle-avoidance behavior.

AdaSR: Adaptive Streaming Reasoning with Hierarchical Relative Policy Optimization

Large reasoning models typically follow a read-then-think paradigm: they observe the complete input, reason over a static context, and then produce the answer. Yet many real-world scenarios are inherently dynamic, such as audio and video stream, where information arrives as a continuous stream and models must reason, update, and respond under partial observations. Recent streaming reasoning methods allow models to think while reading, but they largely rely on supervised imitation of pre-constructed trajectories, which limits their flexibility. In this paper, we propose AdaSR, an adaptive streaming reasoning framework that enables models to reason during input streaming and perform final deliberation once the stream is complete, learning when to think, and how much computation to allocate across different stages. To optimize this hierarchical reasoning process, we introduce Hierarchical Relative Policy Optimization (HRPO), which decomposes policy optimization into streaming reasoning and deep reasoning phases, providing more fine-grained advantage assignment instead of uniformly distributing a single sequence-level advantage over all tokens. HRPO integrates format, accuracy, and adaptive thinking rewards to enforce valid reasoning protocols, preserve final task performance, and encourage latency-aware computation allocation. Experiments show that AdaSR achieves a better balance among reasoning accuracy, computational efficiency, and streaming latency compared with supervised fine-tuning baseline. We release our code at https://github.com/EIT-NLP/StreamingLLM/tree/main/AdaSR.

CompRank: Efficient LLM Reranking via Token-Level Compression and Decoding-Free Scoring

Large language model (LLM) rerankers have become an important component of modern retrieval and retrieval-augmented generation pipelines, but their high computational cost limits their applicability to long candidate lists. In this paper, we propose \textbf{CompRank}, a token-efficient reranking framework that reduces redundant computation by aligning reranker design with the sparsity of ranking signals. CompRank decouples document representations from candidate order and query context, enabling reusable document-side states; applies segment-wise token compression to reduce query--document interaction cost; and introduces a CopyNet-style objective that directly aligns attention-based document scoring with training supervision. Experiments on seven BEIR datasets show that CompRank achieves strong reranking performance while retaining only 10.2\% of document tokens, reaching an average NDCG@10 of 39.2 compared with 39.7 under full-token attention. Further scaling experiments on TREC-COVID show that CompRank remains stable when evaluated on candidate lists of up to 500 documents after training on 30-document lists, while achieving $4.9\times$--$9.5\times$ end-to-end speedup over generation-based listwise reranking and approximately $1.3\times$ speedup over the full-token CompRank variant. These results suggest that token-level compression and decoding-free attention scoring provide an effective path toward scalable LLM-based reranking.

miniReranker: Efficient Multimodal Reranking through Visual Cache Reuse and Interaction Sparsity

Multimodal large language models (MLLMs) have recently shown strong potential as point-wise rerankers by directly modeling query--document relevance through next-token prediction. However, point-wise reranking suffers from substantial repeated computation across query--document pairs, while the causal structure of transformers allows only prefix segments to be reused via pre-caching. To address the misalignment of existing query-first and document-first formats with both VQA-style prompting and computation-aware reuse, we propose a \textit{vision-first} formulation that improves both cache reuse efficiency and reranking performance. However, the remaining cost is still considerable and stems from three main sources: (1) \textit{model depth}, for which we reduce active parameters via early exit; (2) \textit{cross-segment attention}, which we restrict to a narrow interaction band across a few layers; and (3) \textit{visual tokens}, where we reduce the number of tokens via embedder-guided pruning. Together, these designs form miniReranker, which reduces reranking runtime to <1% of the dense implementation under high-reuse settings for a single query, while preserving >96% of the dense model performance.

Beyond FLOPs: Benchmarking Real Inference Acceleration of LLM Pruning under a GEMM-Centric Taxonomy

Pruning has emerged as a dominant paradigm for accelerating large language model (LLM) inference, spanning a broad spectrum of methods that remove computation across tokens, layers, heads, dimensions, and attention patterns. Despite sharing the same objective, these pruning approaches induce fundamentally different execution behaviors, causing realized speedups to depend heavily on hardware and kernel implementations. Consequently, the practical acceleration benefits of different pruning families remain poorly understood. In this work, we introduce a GEMM-centric taxonomy that reorganizes existing pruning methods according to the logical \textbf{M}, \textbf{N}, and \textbf{K} dimensions of general matrix multiplication (GEMM). Leveraging this abstraction, we build a unified benchmarking framework that enables implementation-consistent comparison across the pruning design space and systematically characterizes the acceleration--quality Pareto frontier. Our results show that static depth pruning remains the strongest Pareto-optimal baseline and stays closest to its theoretical acceleration upper bound in memory-bounded scenarios. During prefill, the frontier transitions from static depth at low quality loss (0\%--4\%), to dynamic depth at moderate loss (5\%--16\%), and finally to static width pruning at higher loss levels (17\%--26\%). These findings establish the first unified view of the practical limits of pruning-based LLM acceleration and provide guidance for future pruning research.\footnote{Code is available at https://github.com/EIT-NLP/LLM-Pruning/tree/main/PruningInferSim}

Escaping the KL Agreement Trap in On-Policy Distillation

On-policy distillation (OPD) provides dense token-level supervision by asking a teacher to score student-generated rollouts. However, when the student drifts into an unrecoverable prefix, the teacher may locally agree with the degraded state, producing low reverse KL but little corrective training signal. We identify this persistent regime as a low-KL agreement trap. Further analyses show that tokens during and after such traps produce less useful supervision signals. We propose KAT (KL Agreement Trap Termination), an online OPD termination rule that detects persistent low-KL agreement with a dynamic training-adaptive threshold. By filtering weak supervision from degenerate agreement, KAT improves avg@k accuracy by 2.66% and pass@k by 3.43% across four mathematical benchmarks, while reducing average rollout length by 59.73%.

When Should Queries Be Decomposed? A Stage-Aware Study of Query Decomposition for Multi-Condition Retrieval

Multi-condition retrieval requires systems to identify documents that satisfy multiple distinct constraints, moving beyond mere topical relevance. While query decomposition is widely adopted as an intuitive remedy, its effectiveness across different retrieval pipeline stages remains underexplored. In this paper, we conduct a stage-aware empirical study and uncover a stark, stage-dependent effect: decomposition during initial retrieval frequently harms retrieval performance due to semantic dilution, yet substantially improves reranking by enabling more fine-grained constraint verification. Motivated by these insights, we propose a principled Stage-Aware Decomposition framework that retains the monolithic query during initial retrieval to preserve global semantic context, while employing sub-queries exclusively during reranking for fine-grained constraint matching. Extensive evaluations on the MultiConIR and SSRB benchmarks demonstrate that our framework consistently improves ranking performance for compositional queries across multiple retrieval and reranking models. We release our code at https://github.com/EIT-NLP/Query-Decompose.