Knowledge catalogue
arXiv:2608.25115v1 Announce Type: new Abstract: Existing methods for improving Retrieval-Augmented Generation (RAG) efficiency mainly optimize downstream LLM generation, such as context compression or
I wanted to see just how capable Qwen3.8-27b is locally. I have a RTX 4090 and 96GB of RAM but the Q4 comfortably fits in the GPU with plenty of context, the few times I needed more than 130k context
The Qwen3.8‑Flash‑Next 176 B multimodal MoE model uses a hybrid Gated DeltaNet / Qwen Sparse Attention (QSA) architecture that compresses long‑context history and aggregates tokens into micro‑blocks f
arXiv:2608.23834v1 Announce Type: new Abstract: The key-value (KV) cache is a primary capacity and bandwidth bottleneck in long-context LLM serving. We present Minima-KV, a retention-preserving hierar
arXiv:2608.23962v1 Announce Type: new Abstract: When an LLM serving deployment runs out of KVcache room, there are two well-established ways out. Tensor parallelism shards the weights and the KV cache
Companion to the Qwen 3.8 Release Megathread. Compiled from ~2,000 posts scanned across both subs, with deep reads of the 45 highest-signal threads (560 posts and comments), Aug 15–22, 2026, plus inde
arXiv:2608.18813v1 Announce Type: cross Abstract: In the last years, a number of proofs of the fact that O_2 is a multiple context-free grammar (MCFG) were given. Such results can be exploited in the
arXiv:2608.18108v1 Announce Type: cross Abstract: Large language models are being incorporated into sensitive and important decision-making processes across nearly all fields. While prior work studies
arXiv:2608.17168v1 Announce Type: cross Abstract: Reasoning has become a standard technique and feature for contemporary LLMs; however, its application and quality in the context of demanding legal-or
arXiv:2608.17336v1 Announce Type: new Abstract: Long-context prefill in large language models (LLMs) incurs substantial computation and memory traffic because dense self-attention computes quadratic q
arXiv:2608.14863v1 Announce Type: cross Abstract: LLM-based coding agents have advanced rapidly on single-process SWE tasks, with frontier models now clustering in the high-70s on SWE-bench Verified.
arXiv:2608.13578v1 Announce Type: cross Abstract: Transformer architectures rely on dense self-attention to model long-range dependencies, but this mechanism exhibits quadratic complexity with respect
Thought i would just add my own flags here for llama.cpp (literally pulled and rebuilt latest today). Running on a 4090 FE and 48gb of DDR4 ram on WSL2. Basically you have 3 knobs you can tune for max
arXiv:2608.13921v1 Announce Type: new Abstract: LLM agents increasingly maintain personal memory across sessions, but it can conflict. Preferences depend on context, behavior evolves, and sources can
So, I have been dreaming of getting 17 tokens per second using my RTX 3050 6GB version on a decent context window for Hermes needed above 60k. The hope is that has was a 22GB of DDR 4, hoping they can
arXiv:2602.24287v2 Announce Type: replace-cross Abstract: In multi-turn conversations, large language models typically condition on the full conversation history: both past user prompts and assistant
arXiv:2608.10528v1 Announce Type: cross Abstract: Anchor-based pointwise LLM reranking scores each candidate against a shared reference passage to recover cross-document context at pointwise cost. We
I did some testing this morning, and I was surprised to find that Muse Glimmer actually comfortably fits on a single RTX 3090 with full context + DFlash + mmproj at Q4_K_XL, unlike Qwen3.6-27B and Gem
(I am not a native speaker, written by myself, so please bear with me) I really want to like DeepSeek-V4-Flash-0731. But it has serious flaws that don't align with the high score on intelligence bench
TL;DR: On a Qwen3.6-35B-A3B Q6 setup sized for 64K context on a 24GB RTX 3090, spilling eight MoE expert layers to CPU freed enough VRAM to increase -b from 512 to 1024 and -ub from 128 to 512. Prompt
arXiv:2608.05124v1 Announce Type: cross Abstract: Long context reasoning in large language models (LLMs) is usually constrained by the fact that a single inference trajectory has to simultaneously exp
arXiv:2608.01056v1 Announce Type: cross Abstract: Tool-using language-model agents are governed not only by task prompts but also by persistent system-side instructions that specify tools, arguments,
arXiv:2608.00239v1 Announce Type: new Abstract: Learning transferable representations from CT-report pairs requires combining whole-volume context with anatomy-specific evidence. Existing methods typi
AI agents are only as good as the instructions and context you give them. When we launched Google Agent Skills, our goal was simple: encode Google Cloud domain knowledge into structured, open-source i
If you maintain an AGENTS.md or a CLAUDE.md, this is worth a read. (bookmark it) 288 gold-test evaluated runs across Claude Code and Codex, 17 real tasks from 3 repositories, with context-injection st
arXiv:2607.26520v1 Announce Type: cross Abstract: Conversational AI agents commonly lack persistent memory across sessions. The obvious fixes like injecting full chat histories into the context window
Welcome to the second Cloud CISO Perspectives for July 2026. Today, Chris Betz, CISO, Google Cloud, and Alicja Cade, Senior Director, Office of the CISO, Google Cloud, explain what boards of directors
arXiv:2607.27090v1 Announce Type: cross Abstract: Large language models are increasingly deployed with persistent personalized context, such as accumulated memory profiles or long conversation histori
Figure 1: CUDA-to-MLX optimization translation map. CUDA optimization knowledge can be translated into architecture-native MLX strategies rather than copied instruction-for-instruction. We face a new
arXiv:2607.23175v1 Announce Type: cross Abstract: Reducing toxicity is often framed as a global alignment problem, yet perceptions of harmful language are subjective and context-dependent. We present
arXiv:2607.22646v1 Announce Type: new Abstract: Large language models (LLMs) display a striking ability to predict next observations from Hidden Markov Models (HMMs) via in-context learning (ICL), but
arXiv:2607.23933v1 Announce Type: cross Abstract: As LLM agents increasingly rely on the Model Context Protocol (MCP) to invoke isolated external sandboxes, disaggregated sandbox deployment introduces
The math behind reinforcement learning (RL) post-training for large language models (LLMs) is notoriously unforgiving. As frontier AI labs push the boundaries of reasoning and coding models using RL p
Artificial intelligence optimization startup Refiant Inc. today launched Protea, a suite of long-context AI models led by a 10 million-token context window that the company says ranks among the larges
arXiv:2607.01002v1 Announce Type: cross Abstract: In long-context use, large language models frequently synthesize answers from the meaning of a relevant context span rather than literally copy-pastin
arXiv:2606.30709v1 Announce Type: cross Abstract: Hierarchical Global Attention (HGA) is a drop-in replacement for dense causal attention in pretrained long-context transformers. HGA preserves the ori
arXiv:2606.07878v1 Announce Type: new Abstract: The KV cache is the memory bottleneck of long-horizon language model deployment. Practically, a deployable compactor must be lightweight enough to call
Google Cloud Storage (GCS) is a foundational component of the modern agentic tech stack and the preferred home for unstructured data at scale. As enterprises deploy agents in production, the critical
arXiv:2605.22082v1 Announce Type: new Abstract: We present CoRMA(Contrastive Robotic Motor Adaptation), a context-based meta-adaptation framework that modifies RMA for force-dominant assembly. CoRMA r
arXiv:2605.18833v1 Announce Type: cross Abstract: Automated data quality assessment is crucial for managing big data, but existing solutions face challenges in achieving accurate context-aware assessm
arXiv:2605.19049v1 Announce Type: cross Abstract: Linear attention has recently gained significant attention for long-context inference due to its constant decoding cost with respect to context length
arXiv:2605.18635v1 Announce Type: cross Abstract: Credit default prediction is a tabular learning problem with severe class imbalance, heterogeneous features, and tight latency budgets. Tabular Founda
arXiv:2605.17625v1 Announce Type: new Abstract: As Large Language Models (LLMs) evolve into persistent scientific collaborators, context window saturation has emerged as a critical bottleneck. Scienti
arXiv:2602.03922v3 Announce Type: replace Abstract: Standard sequence mixing layers used in language models struggle to balance efficiency and performance. Self-attention performs well on long context
Ollama offers several memory management settings, including the ability to reduce context length (context window) to decrease VRAM usage, which can save 0.2-0.4GB for 7-8B models when reduced from 8K
As AI coding agents become deeply embedded in developer workflows, defenders must evolve their definition of malicious files and rethink how to protect against them. Autonomous AI agents operate acros
arXiv:2605.10810v1 Announce Type: new Abstract: We introduce an automatically generated benchmark for predicting hidden text in technical papers. A paper supplies visible context X and a hidden contin
arXiv:2605.09778v1 Announce Type: cross Abstract: Evaluating softmax attention over a fixed long context requires reading every cached key-value pair for each new query token. For a given context (a b
arXiv:2605.04039v1 Announce Type: new Abstract: Clinical LLMs are often scaled by increasing model size, context length, retrieval complexity, or inference-time compute, with the implicit expectation
arXiv:2605.02743v1 Announce Type: cross Abstract: The field of sensor-based human activity recognition (HAR) mainly uses posture, motion and context data of Inertial Measurement Units (IMUs) to identi
At Google Cloud Next ‘26, we announced that more than 50 Google-managed Model Context Protocol (MCP) servers are generally available or in preview, with more on the way. Why it matters: To move beyond
arXiv:2604.20915v1 Announce Type: cross Abstract: Transformers suffer from a high computational cost that grows with sequence length for self-attention, making inference in long streams prohibited by
Highlights: 👉 SOTA coding—93.5% LiveCodeBench, Codeforces 3206, and 80.6% SWE-Bench Verified 👉 Hybrid attention efficiency—27% FLOPs and 10% KV cache vs V3.2 for long-context inference 👉 Three reasoni
arXiv:2604.21816v1 Announce Type: new Abstract: The Model Context Protocol (MCP) has become a common interface for connecting large language model (LLM) agents to external tools, but its reliance on s
arXiv:2604.21144v1 Announce Type: cross Abstract: Situated dialogue requires speakers to maintain a reliable representation of shared context rather than reasoning only over isolated utterances. Curre
arXiv:2604.19777v1 Announce Type: cross Abstract: Large Language Models (LLMs) exhibit a well-documented positional bias when processing long input contexts: information in the middle of a context win
At Google Cloud Next, we are announcing innovations across every layer of our storage stacks — performance, intelligence, and management — to ensure your data is as fast and as useful as the AI models
arXiv:2602.01566v2 Announce Type: replace Abstract: Deep research is emerging as a representative long-horizon task for large language model (LLM) agents. However, long trajectories in deep research o
arXiv:2604.14170v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) grounds Large Language Models (LLMs) in external knowledge but often suffers from flat context representations and
arXiv:2601.12145v2 Announce Type: replace Abstract: Softmax attention struggles with long contexts due to structural limitations: the strict sum-to-one constraint forces attention sinks on irrelevant