Nexus introduces a depth‑adaptive KV‑cache splicing and retrieval‑decoupled tool routing mechanism for agentic large language models that reduces the time‑to‑first‑token (TTFT) by decoupling tool routing from the expensive schema re‑encoding step. It uses an INT8 semantic lookaside buffer to select tools via retrieval and generates arguments from a compressed textual signature, maintaining about 89% routing accuracy even as the tool registry scales to 250 tools. Additionally, Nexus can splice compiled schema KV blocks into the live context, repairing the seam with a depth‑adaptive suffix redecode when rotary position embedding drift exceeds a threshold, ensuring output fidelity while achieving up to 1.7× TTFT speedup at moderate depth.
arXiv:2606. 01502v1 Announce Type: cross Abstract: Frontier LLMs increasingly decide what a query attends to with a sparse-attention indexer that picks a few KV-cache blocks per query: attention's unit is now a small, reusable chunk.
arXiv:2607. 18553v1 Announce Type: cross Abstract: Can a language model read the quality of ongoing computation, and can an external intervention turn that readout into better outcomes?
arXiv:2608. 09942v1 Announce Type: cross Abstract: It is widely assumed that chain-of-thought (CoT) prompting universally improves LLM reasoning.
The paper investigates whether training Mixture-of-Experts (MoE) routers can improve memory‑bandwidth locality on consumer GPUs. Using a new zero‑surgery telemetry tool, the authors measure that a large Qwen3‑235B model is bottlenecked by disk‑based expert access, and that an LRU cache can serve a majority of requests. They pre‑register experiments training 137 M‑parameter MoE models with locality‑aware losses, finding that while cache misses can drop up to 60 % (99 % static‑pin hit rate), every configuration fails to meet a strict 1 % perplexity threshold, indicating a tight coupling between cache efficiency and model quality.
arXiv:2606. 24467v1 Announce Type: new Abstract: Long-context large language model (LLM) inference is increasingly constrained by the memory footprint and decoding cost of key-value (KV) caches, limiting sustainable deployment on resource-constrained hardware.
The paper introduces XKV, a latent protocol that enables efficient communication between heterogeneous language models by translating a sharer's key‑value cache into a receiver's context. XKV overcomes limitations of prior methods by jointly pooling both caches, reconciling differing layer depths, and allowing each receiver position to retrieve its own residual in native KV geometry. Across 45 dataset‑model pairings, XKV outperforms previous protocols and text communication while using fewer parameters and achieving faster translation times.
arXiv:2606. 09659v1 Announce Type: cross Abstract: Long-context language model inference is bottlenecked by memory, as the KV cache grows with context length.
arXiv:2606. 09916v1 Announce Type: cross Abstract: Multi-turn LLM agents fan short queries into long trajectories of tool calls, search results, and intermediate reasoning.
Long-context large language model (LLM) inference is increasingly constrained by the memory footprint and decoding cost of key-value (KV) caches, limiting sustainable deployment on resource-constrained hardware. Existing KV cache eviction methods typically apply heuristic token scoring over all heads in GQA-based LLMs.
Serving a 235B-parameter Mixture-of-Experts (MoE) model on a single 8 GB GPU is bottlenecked not by compute but by memory bandwidth: decode must stream each token's active experts from whichever tier holds them, and on consumer hardware most experts sit on an SSD far slower than RAM. We quantify this bandwidth wall on Qwen3-235B (Q4_K_M, 134 GB): measured decode is 0.
arXiv:2608. 02560v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) imposes a prefill cost proportional to retrieved context length, and -- with Transformer backbones -- a KV-cache that grows with each generated token.
Long-context language model inference is bottlenecked by memory, as the KV cache grows with context length. Recent techniques to compress the KV cache fall short: they either degrade model quality substantially or require considerable time and compute to compress a single long prompt.