Radar: Meta storage architecture
A July 5, 2026 publication describes a storage architecture and performance techniques, including data locality, cache unification, and pre-fetching.
Read articleKnowledge you can apply
Radar
Daily bookmarks: relevant news, facts and discoveries.
EfficientRollout aims to speed up RL rollouts
The paper describes system-adaptive self-speculative decoding for RL rollouts in LLMs. Its available summary reports up to 19.6% faster generation; another figure is cut off.
Read articleHASTE organizes skills for ML engineering agents
Published on July 5, 2026, the entry describes HASTE, a hierarchical multi-agent system. In an ablation with 159 skills across eight competitions, hierarchical loading achieved a 100% medal rate, compared with 62.5% for flat loading.
Read articleHOLA adds a 64-token cache to linear attention
A post describes a parameter-free token-surprise metric and a 64-token cache intended to address catastrophic forgetting in linear attention. It claims to outperform full-attention baselines.
Read articleCosine taper lowers perplexity without expanding the budget
The paper examines parameter allocation across layers and reports lower perplexity with a cosine taper across several architectures and scales, without increasing parameter count or FLOPs.
Read articlePruning or training small LLMs from scratch
A paper compares six pruning methods for Llama-3.1-8B with training from scratch under two token-budget scenarios. The advantage depends on how the budget is counted and on pruning granularity.
Read articleLocal branching to scale LLMs at inference time
Published on July 5, 2026, the material presents a local branching approach that previews possible next tokens and selects a path before generation continues.
Read articlePost-training RL: tuning one layer
A post about a paper reports that reinforcing high-contribution intermediate layers may match or outperform RL over all parameters while changing fewer parameters.
Read articleRubricEM uses rubrics to guide research agents
Published on July 5, 2026, the item describes a method that trains agents to create rubrics and use them for planning, searching, reviewing, and answering.
Read articleCornell offers a workshop on GPU architecture
The Cornell Virtual Workshop provides introductory material on GPU architecture, a subject that can help engineers analyze GPU performance.
Read articleArticles and talks on GPU optimization
A collection of Vasily Volkov’s work on GPU performance, covering topics such as loop unrolling, latency hiding, dense linear algebra, and GPU memory.
Read articleReward signals can diverge from human intent
Published on July 4, 2026, the post describes four signals for evaluating coding agents and warns that they can drift from what people intend.
Read articleFlowTracer tracks information to guide RL
The article presents a method that uses attention signals to trace information flow in language-model reasoning and identify influential tokens. Those signals can guide credit assignment in reinforcement learning.
Read articleUnderstanding latency hiding in GPUs
A chapter by Vasily Volkov of UC Berkeley covers how GPUs hide latency and introduces concepts in GPU architecture and performance.
Read articleGPU Execution Model: a chapter for study
A chapter in Modern GPU Programming for MLSys presents the GPU execution model as a study topic for people working with modern programming and machine-learning systems.
Read articleTinyRouter tests routing across open models
Published on July 4, 2026, the account describes a router with about 10,000 parameters and initial benchmark results, while noting that more experiments are needed.
Read articleHOLA proposes an exact KV cache for linear attention
Published on July 3, 2026, the article summary says HOLA combines a recurrent state based on the delta rule with a bounded, exact key-value cache. It reports improved Wikitext perplexity and robust needle recall on RULER up to 32k tokens.
Read articleStudy examines in-context retrieval with millions of tokens
Published on July 3, 2026, the report describes a study of answer retrieval from large in-context corpora and generalization to longer lengths.
Read articleHow Statically Linked Programs Start on Linux
Published on July 2, 2026, Eli Bendersky's piece explains how statically linked programs run on Linux, focusing on what happens during program startup.
Read articleOpen-source real-time voice demo
Published on July 2, 2026, the briefing describes a real-time voice demo with open-source models and code, supporting WebSocket or WebRTC.
Read articleRota Nacional
Bring privacy into your workflow.
30 days, no card, with a starting quota. After that, Pix credit from R$ 10,00.