Large vision-language models (LVLMs) remain highly sensitive to text embedded in images: a misleading word printed, painted, or pasted onto an object can override visual evidence and flip the model’s answer. Existing defenses are often tied to specific vision encoders, require training or feature-level intervention, or do not operate on the autoregressive decoding interface of modern LVLMs. We propose TypoCD, a training-free contrastive decoding method for typographic robustness. TypoCD constructs a text-localized counterfactual view by detecting visible text regions and inpainting only the typographic content, then contrasts the attacked-image and typography-removed logits during decoding. This localized contrastive direction targets the attacker’s textual signal while preserving unrelated visual content, allowing TypoCD to subtract the typography-induced logit shift without modifying model parameters or internal features. Across three open-source LVLM backbones and synthetic, real-world, and safety-oriented typographic benchmarks, TypoCD recovers attacked-image accuracy to within a few percentage points of clean-image performance without per-backbone tuning, closing a gap that often exceeds 40 percentage points under attack. When combined with a simple prompt prefix, TypoCD also substantially reduces jailbreak attack success on safety-oriented typographic attacks.

Large Language Models (LLMs) are often augmented with Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT) prompting, yet static “always-on” use is computationally wasteful. Existing adaptive methods typically optimize a single axis, overlooking that evidence need and reasoning depth are only partially correlated. We present , a dual-axis routing framework that separates retrieval necessity from reasoning necessity under a user-defined cost–quality trade-off. Offline, profiles four pipelines (Direct, RAG, CoT, RAG+CoT) and derives supervision by selecting the utility-maximizing strategy that trades answer quality against token usage and latency. Online, a compact dual-head router, conditioned on cost weights, uses lightweight probes—retrieval-score dispersion (NQC) and single-pass draft negative log-likelihood (NLL)—to decide whether to invoke RAG and/or CoT without sampling or model internals. Across six QA benchmarks, reduces token usage by up to 86% and latency by up to 84% while improving answer quality over strong baselines.

Large Language Models (LLMs) promise impressive capabilities, yet their multi-billion parameter scale makes on-device or low-resource deployment prohibitive. Mixed precision quantization offers a compelling solution, but existing methods struggle when the average precision drops below four bits, as they rely on isolated, layer-specific metrics that overlook critical inter-layer interactions affecting overall performance. To address these limitations, we first frame the mixed-precision quantization problem as a cooperative game among layers and introduce Shapley-based Progressive Quantization Estimation (SPQE) to efficiently obtain accurate Shapley estimates of layer sensitivities and inter-layer interactions. Leveraging the SPQE estimates, we propose Cooperative Game Inspired Mixed-Precision Quantization (CoopQ) which translates these Shapley estimates into a binary quadratic optimization formulation, assigning either 2 or 4-bit precision to layers under strict memory constraints. Comprehensive experiments conducted on Llama-3, Gemma-2, and Qwen models across three independent PTQ backends (Quanto, HQQ, GPTQ) demonstrate CoopQ’s scalability and consistently superior performance compared to methods relying solely on isolated metrics. Across average precisions spanning 4 bit down to 2 bit, CoopQ cuts Perplexity by 20 – 80 % relative to the best baseline, with the margin growing as the bit-width tightens.

Graph-based Retrieval-Augmented Generation (GraphRAG) can fail in interdisciplinary medicine where evidence comes from incommensurable knowledge systems, e.g., Traditional Chinese Medicine (TCM) and Western Medicine (WM). Similarity-driven retrieval in a shared latent space may over-unify heterogeneous concepts, causing semantic collapse (spurious alignments, provenance-blind mixing, and broken multi-hop chains). We propose Xonto-RAG, a cross-ontology GraphRAG framework that avoids ontological unification by constructing a dual-ontology graph with typed bridges induced via LLM-assisted verify-and-vote. A query-aware Relational Graph Attention Network (RGAT) then assembles compact, provenance-tracked multi-hop evidence subgraphs for grounded generation. Across TCM–WM benchmarks, Xonto-RAG outperforms strong RAG/GraphRAG baselines, seeing up to +22.2% accuracy on MLEC-QA TCWM (vs. OG-RAG), 77.0% accuracy with a 7B backbone, and 85.1% faithfulness on MLEC-TCM, while reducing semantic-collapse failures and yielding interpretable reasoning chains.