DeCoRAG: Cognitive Decoupling and Semantic-Aware Cropping for Complex Document Understanding

Advancing multimodal retrieval-augmented generation (RAG) for complex document understanding presents a formidable dual dilemma of accuracy and efficiency, particularly in graph RAG. Processing structurally sparse yet visually dense layouts, such as extracting a tiny data marker from a financial chart, often incurs computationally prohibitive token overhead while still triggering catastrophic hallucination. However, multimodal Graph RAG pipelines rely on graph-construction stages that assume Vision-Language Models (VLMs) can resolve sparse semantics within high-density layouts. We challenge this assumption, revealing that forcing VLMs to localize visual evidence, interpret semantics, and extract relations triggers a "Visual Attention Sink," a mechanism driving catastrophic semantic loss, while full-page processing incurs massive computational overhead. Controlled interventions verify that this failure is boundary-driven rather than content-specific and that semantic anchoring mitigates it. To fundamentally correct this flawed paradigm, we introduce DeCoRAG, a multimodal Graph RAG pipeline that shifts knowledge processing from coupled visual-semantic reasoning to "Cognitive Decoupling." Rather than passively processing raw pixels, its graph-construction stage establishes a macroscopic Semantic Anchor to neutralize the attention sink. This anchor subsequently drives our Region-Aware Pruning and Cropping (RAP-Crop) mechanism, shifting the reasoning space from dense, noisy backgrounds to purified, intent-driven semantic clusters. The resulting graph supports hybrid retrieval and answer generation. Across complex document benchmarks, DeCoRAG improves the semantic pass rate by up to 12.5 percentage points over the strongest baseline and generalizes to DocVQA. RAP-Crop reduces offline graph-construction prompt tokens by 40.8% without sacrificing end-to-end accuracy.

KAP: Bridging the Knowledge Selection-Runtime Consumption Gap in LLM Systems

Modern LLM systems increasingly rely on knowledge-selection processes that produce high-value structured priors, such as ranked evidence, graph topology, multimodal alignment, and confidence signals. Yet LLM serving remains fundamentally oblivious to this rich structure: once such signals are serialized into a prompt, the backend observes only a flat token sequence, forcing dense and uniform consumption of the full key-value (KV) state during decoding. We term this architectural mismatch the Knowledge Selection-Runtime Consumption (KSRC) gap: richer contexts enlarge the full-prompt KV footprint and decode-time memory traffic, increasing latency and degrading throughput even when reasoning depends on only a small fraction of the context. To bridge the gap, we propose Knowledge Access Planning (KAP), a paradigm-shifting execution abstraction that elevates structured knowledge priors from passive prompt-construction hints into first-class physical execution artifacts. KAP establishes a universal intermediate representation (IR)-the runtime access plan-which compiles structured knowledge signals to govern physical KV access without altering logical prompt semantics, model weights, or training procedures. Through this IR, KAP shifts LLM serving from token-aware context consumption to plan-driven, knowledge-aware runtime consumption. We instantiate KAP with GraphSpec, a compiler-executor realization connecting structured knowledge selection to an LLM serving backend. We derive a phase-boundary model for the positive-speedup regime of plan-guided execution. Across 4K-128K long-context QA workloads, GraphSpec maintains answer quality comparable to full-context decoding while decoupling physical KV consumption from prompt length, reducing proposal-time KV access to 5.5% of source KV state at 128K, and fundamentally shifting the scaling trajectory of long-context generation.

OrchANN: A Unified I/O Orchestration Framework for Skewed Out-of-Core Vector Search

Approximate nearest neighbor search (ANNS) at billion scale is fundamentally an out-of-core problem: vectors and indexes live on SSD, so performance is dominated by I/O rather than compute. Under skewed semantic embeddings, existing out-of-core systems break down: a uniform local index mismatches cluster scales, static routing misguides queries and inflates the number of probed partitions, and pruning is incomplete at the cluster level and lossy at the vector level, triggering "fetch-to-discard" reranking on raw vectors. We present OrchANN, an out-of-core ANNS engine that uses an I/O orchestration model for unified I/O governance along the route-access-verify pipeline. OrchANN selects a heterogeneous local index per cluster via offline auto-profiling, maintains a query-aware in-memory navigation graph that adapts to skewed workloads, and applies multi-level pruning with geometric bounds to filter both clusters and vectors before issuing SSD reads. Across five standard datasets under strict out-of-core constraints, OrchANN outperforms four baselines including DiskANN, Starling, SPANN, and PipeANN in both QPS and latency while reducing SSD accesses. Furthermore, OrchANN delivers up to 17.2x higher QPS and 25.0x lower latency than competing systems without sacrificing accuracy.