Search: “context”

TL;DR: Every public low-bit GGUF of this model is secretly ~4.70 bpw. Shim the rows to 256 and it becomes a real 3.07 bpw / 11.77 GiB file that runs 262K context on 16GB. Needs patched llama.cpp — not

arXiv:2608.26140v1 Announce Type: new Abstract: Diffusion Large Language Models (DLLMs) enable non-autoregressive decoding and bidirectional context modeling, but efficient inference remains challengi

arXiv:2608.26199v1 Announce Type: new Abstract: We ask whether AI agents powered by locally deployed large language models can reliably automate expert-defined hardware design workflows in an industry

Hey, An LLM writing into durable memory is an unreliable author. It invents, mislabels, restates, and contradicts. Once one of those mistakes is saved, the agent can carry it into every conversation t

arXiv:2608.26866v1 Announce Type: new Abstract: Many multimodal tasks depend on how visual elements are ordered and composed, not only on recognizing them in isolation. Internet memes are a compact ca

arXiv:2608.26180v1 Announce Type: cross Abstract: Shopping assistants are shifting from ranked product lists toward structured decision support, where systems must synthesize shopper context, product

I've got Qwen3.8-Flash-next running on RTX 3090, Ryzen 9 3950X, a PCIe 3.0 motherboard, and 64GB DDR RAM from 2020. IQ4_XS weights, full kvarn5 context, vision on GPU, experts in host RAM, n-grams on

Creative teams produce more assets than ever, but fragmented tools and manual context transfer slow production. This post shows how to build a reusable agent harness with Amazon Quick and fal, connect

arXiv:2608.25894v1 Announce Type: new Abstract: Large language models (LLMs) frequently produce confident yet factually incorrect responses when user inputs contain misleading premises, a phenomenon w

Quantising the K/V cache is the highest-leverage memory setting most people never touch, and it has one behaviour in the docs that catches almost everyone. Every number below is out of Ollama's own FA

arXiv:2608.25585v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models provide a versatile foundation for general robotic manipulation, yet they exhibit significant brittleness when confr

Presented by EDB As enterprises give AI agents more autonomy — the ability to plan, decide, and act across systems without a human approving each step — a hard question moves to the center of every ar

Qwen3.8‑Flash‑Next is a multimodal mixture‑of‑experts model with 125 B parameters that employs Gated DeltaNet (compressing historic context) and Qwen Sparse Attention (exact retrieval over full contex

In the past, I have use llama.cpp, but I read that the exl3 quantization format should give better precision, so I have tried exllamav3/tabbyAPI. It was able to write the shown simple HTML game withou

arXiv:2608.23965v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) improves the factuality of large language models by grounding responses in external documents, but it also expose

arXiv:2608.23908v1 Announce Type: new Abstract: Tax-loss harvesting demonstrates consistent benefits to long-term portfolio growth; yet implementing it efficiently often involves complex consideration

arXiv:2608.23953v1 Announce Type: cross Abstract: An agent harness is what turns a language model into an autonomous agent: the surrounding code that builds the model's context, mediates its tools, ru

arXiv:2607.18467v2 Announce Type: replace Abstract: Many operational decisions rely on predictive models that estimate uncertain outcomes conditional on observable contexts. Training such models, howe

arXiv:2608.24022v1 Announce Type: cross Abstract: LLM agents integrated with external resources gain complex task capabilities, yet the unified natural-language context channel makes them vulnerable t

arXiv:2608.23026v1 Announce Type: cross Abstract: Multilingual LLM outputs can vary across sociocultural contexts. However, evidence of cultural grounding can be misleading: identity labels may be inf

arXiv:2608.22446v1 Announce Type: new Abstract: Metaphors are figurative use of words for conceptual mapping. Metaphor detection in the legal context has been crucial as metaphors are persuasive jurid

Impressive work from Alibaba. (bookmark it) If you build long-running agents and keep rewriting your memory schema, take a look at this approach. It basically treats agent context management as a prog

arXiv:2608.22731v1 Announce Type: new Abstract: Nonverbal behavior generation systems for virtual agents often take an utterance as input and generate nonverbal behaviors that emphasize or illustrate

My best bet is the embedding-offloaded linear where the 51b n-grams track semantics and context, like 3.8 with its loss of real world knowledge bolted back on. Otherwise? Sparse full-attention with de

v4.5.0: Added native desktop control, agent execution budgets, concurrency limits, loop protection, and safer local embedding fallback. v4.6.0: Rebuilt agents around durable parent-led orchestration.

arXiv:2608.22752v1 Announce Type: new Abstract: A safety rule and an episodic log compete for the same tokens in an AI agent's context. When the budget overflows, both are summarized at the same rate;

AgentX 1.0 is the first fully open‑source multi‑turn agentic coding inference benchmark (Apache 2.0), released August 2026 with a 1 million‑token context and over 1 million tokens of dataset built at

Hi, Back in late march I begun playing around with Qwen 3.6 27b and found like everyone else that it's notoriously good at tool calls, where every model I tried before just derailed after a few turns

arXiv:2608.21199v1 Announce Type: new Abstract: Transformer-based tabular foundation models such as TabPFN achieve strong predictive performance but incur quadratic computational cost with context len

arXiv:2608.20631v1 Announce Type: new Abstract: Large language model (LLM) agents have demonstrated the ability to solve multi-step tasks requiring planning, tool use, and external information access,

Using llama.cpp I seem to be unable to get my to GPUs working tougether correclty, so I need help somehow. Setup: 96GB RAM, one Blackwell 5000 (48GB) and one 3090 (24GB). I am trying to run the UD-Q3_

arXiv:2608.19659v1 Announce Type: new Abstract: Choosing tensor-parallel (TP) degrees and replica counts for an LLM serving fleet is difficult because performance is not monotonic in TP and the feasib

arXiv:2608.19355v1 Announce Type: cross Abstract: Educational visual question answering, or VQA, requires models to solve curriculum-oriented multiple-choice questions using both language and visual e

I kept running into the same wall. Every agentic coding plugin ships an autocomplete, and almost all of them lock it down — no model choice, or a subscription, or both. I don't want a chat panel, I do

I used LM Studio Bionic with Qwen 3.8 27B Q3_K_S with 57k context. It took a staggering 63 hours to finish coding. After the first prompt 'Create a beautiful, relaxing flight simulator in a single HTM

QwQ was genuine next-gen performance usable on local hardware, but the massive required context (it's reasoning style was akin to 'if I say every possible word, I'll notice the right one!') kinda made

arXiv:2608.19598v1 Announce Type: cross Abstract: Direct Preference Optimization (DPO) has emerged as an effective approach for aligning large language models (LLMs) with human preferences. However, i

arXiv:2603.23522v2 Announce Type: replace-cross Abstract: Evaluating large language models (LLMs) on open-ended questions is difficult because response quality depends on the question's context. Binar

arXiv:2608.19556v1 Announce Type: cross Abstract: Streaming autoregressive diffusion models enable real-time, long-horizon video generation, but their training objectives optimize local frame predicti

arXiv:2608.20024v1 Announce Type: new Abstract: District heating energy hubs require reliable heat load forecasts for efficient operational scheduling. Conventional forecasting workflows train system-

Just got the setup dialed in yesterday. Getting really good results and is making 3.8 usage feel faster in hermes. I got Qwen to specifically use Ling tiny for simple tasks like context compression an

arXiv:2608.19147v1 Announce Type: cross Abstract: Modern Intel AI PCs ship capable integrated GPUs and NPUs with 16+ GB of unified memory, and they spend considerable time idle. That is not enough mem

Component Validated configuration Motherboard ASRock Rack SPC621D8U-2T/OVH CPU Xeon Gold 6330 (Get gold/platinum if interested in Optane Pmem gimmicks) GPU fabric Two Broadcom/PLX PEX88096 islands, ei

arXiv:2608.17843v1 Announce Type: cross Abstract: Large language models (LLMs) have demonstrated strong performance on structured reasoning tasks, but what they encode and whether it informs model beh

arXiv:2608.17827v1 Announce Type: new Abstract: Public institutions face a persistent challenge in selecting LLMs suited to their specific context. Existing benchmarks, however, are of limited use as

Edit: Title says 134 tps, it's actually 138 -- keep in mind my 3090 is power limited to 250w. Three days ago I released a hyper-optimized Qwen3.8-27B inference engine for an RTX 3090 (82 tps single re