Efficient multimodal inference is increasingly constrained not only by model quality or FLOP count, but also by the cost of preserving, moving, routing, caching, and quantizing multimodal representations under latency, memory, and energy constraints. This paper reviews recent advances in efficient vision-language and multimodal large language models, covering visual token compression, video token management, KV-cache optimization, Mixture-of-Experts (MoE) routing, low-bit quantization, edge deployment, and hardware-aware benchmarking.
arXiv:2607. 20981v1 Announce Type: new Abstract: Efficient multimodal inference is increasingly constrained not only by model quality or FLOP count, but also by the cost of preserving, moving, routing, caching, and quantizing multimodal representations under latency, memory, and energy constraints.
arXiv:2609.38823v1 Announce Type: new Abstract: Multimodal mixture-of-experts (MoE) models combine sparse expert activation with visual-language capabilities, yet their inference remains costly becau...
The paper introduces Q-TOFC, a query‑guided task‑oriented visual feature compression method that uses residual vector quantization to encode merged features as compact codebook index sequences. By incorporating query relevance into feature aggregation and adding a quantization error compensation adapter, Q‑TOFC reduces visual payload by 53.6% compared to previous TOFC while preserving task performance. Experiments across seven multimodal benchmarks and latency tests confirm its effectiveness under bandwidth‑constrained uplinks.
arXiv:2512. 08240v2 Announce Type: replace-cross Abstract: Vision-language models (VLMs) rely on hundreds of visual tokens, leading to high computational and memory costs.
arXiv:2609.38090v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) models are a compelling architecture for scaling model capacity, making them especially attractive for deployment on resource-...
The paper introduces Visual Token Coding (VTC), a token compression method for video multimodal large language models that mimics classical video coding by predicting I/P frames and measuring residuals to reduce token redundancy. VTC is extended with dynamic features—Dynamic Resolution Input, Dynamic Token Allocation, and Spatial Coverage Top‑K—forming VTC_Dy, which can be applied to existing MLLMs without additional tuning. Experiments on three MLLMs and multiple video benchmarks show that VTC_Dy retains over 100% of average performance with a 50% token budget and 97.8% with a 25% budget, while the code is publicly available.
The paper introduces RAMP, a method for robust adaptive mixed‑precision quantization of vision models on edge CPUs. It evaluates 13 sensitivity metrics across four neural networks, finding that Jensen‑Shannon Divergence consistently identifies layers that can be safely quantized. Using K‑Means clustering on these metrics, RAMP achieves near‑lossless accuracy with an average 1.81× speed‑up, while cautioning against excluding low‑speed‑up layers that can fragment the computational graph.
The paper introduces TopoCompress, a token compression framework designed for distributed edge Mixture-of-Experts (MoE) inference. It jointly optimizes token compression, expert deployment, GPU-CPU residency, and routing to reduce cross-server communication and resource usage. The method uses a two-timescale alternating optimization, with an online loop compressing low-importance tokens and an offline loop updating expert placement based on accumulated traffic.
arXiv:2606. 17118v1 Announce Type: cross Abstract: Mixture-of-Experts Multimodal Large Language Models (MoE-MLLMs) offer remarkable performance but incur prohibitive GPU memory costs, making compression essential.
arXiv:2606.06158v2 Announce Type: replace Abstract: Adaptive video tokenisation seeks to dynamically allocate token budgets based on the underlying visual complexity of a sequence. Current continuous...
MoE-ViE introduces a Mixture-of-Experts vision encoder that scales efficiently for image and video understanding, outperforming dense counterparts across various sizes. The study shows fine‑grained MoE topologies provide significant gains, and proposes an auxiliary‑loss‑free balancing variant and a specialized MoE kernel to reduce inference latency. With frame‑level distillation and a novel freezing mechanism, the largest MoE‑ViE model matches state‑of‑the‑art zero‑shot performance while being 1.7× larger and 76% faster, and it outperforms other encoders when paired with a language model on both image and video benchmarks.
arXiv:2608.29291v1 Announce Type: new Abstract: Unified multimodal models jointly support understanding and generation, but incur substantial redundant computation across tokens, layers, and generati...