TP

TP-7 Field Recorder

Scaling LLM Inference Beyond Amdahl`s Limits via Eliminating Non-Scalable Overheads

arXiv:2606.01927v1 Announce Type: new Abstract: Deployers of online LLM services usually seek to maximize cluster-wide performance given a fixed number of GPUs. Tensor parallelism (TP) is necessary to fit modern models but scales sub-linearly as the TP degree t grows, due to cross-GPU communication and non-scalable runtime work, as predicted by Amdahl's Law. Conversely, increasing t improves memory efficiency and alleviates KV-cache contention and swapping.

Deterministic Inference across Tensor Parallel Sizes That Eliminates Training-Inference Mismatch

arXiv:2511.17826v2 Announce Type: replace Abstract: Deterministic inference is increasingly critical for large language model (LLM) applications such as LLM-as-a-judge evaluation, multi-agent systems, and Reinforcement Learning (RL). However, existing LLM serving frameworks exhibit non-deterministic behavior: identical inputs can yield different outputs when system configurations (e.g., tensor parallel (TP) size, batch size) vary, even under greedy decoding. This arises from the...

When Do Fewer Coordinates Suffice in DP-SGD?

Announce Type: new Abstract: Differentially private stochastic gradient descent (DP-SGD) injects noise into every updated coordinate, making the injected noise energy scale with the ambient parameter dimension \(d\). We ask when private training can update fewer coordinates without losing the signal needed for optimization. We propose \textsc{TP-TopK} (Two-Phase TopK DP-SGD), a two-phase method for coordinate-sparse private training without public data, in which a private warm-up phase...

US adds popular Chinese firms to Pentagon's 'Chinese military company' list: Who's on it?

The Pentagon on Monday reinstated and expanded a list of Chinese companies it accuses of supporting China's military, naming major firms including Alibaba Group Holding, Baidu Inc., Tencent Holdings and electric vehicle maker BYD Co. The move is likely to add further strain to already tense US-China relations. The US Department of Defense said the companies had been added to its so-called 1260H list, which identifies entities determined by Washington to be aiding China's People's Liberation...

Wi-Fi Router vs. Mesh System: Which Is Best for You?

We all want a good, strong Wi-Fi signal throughout our homes, but figuring out how to achieve it without overspending can be tricky. Even once you’ve decided to replace the router from your internet service provider, do you get a new Wi-Fi router or invest in a mesh system? I covered this briefly in my router buying guide, but I’m going to dive a bit deeper here.

Multi-feature Classification to Improve Colorimetric Loop-Mediated Isothermal Amplification Fidelity

Loop-mediated isothermal amplification (LAMP) is a cost-effective and portable assay technique for performing nucleic acid-based diagnostics in the field whose adoption is hindered by design and reproducibility issues. This is due to a complex primer design process that fine-tunes parameters across 6-8 binding regions. The likelihood of assay success depends on satisfying thermodynamic and secondary structure constraints while maintaining target specificity and avoiding overlaps between...

Unraveling the Mechanism of HIV-1 Hypersusceptibility to Tenofovir Imparted by Islatravir Resistance Mutations

In response to the newly approved antiretroviral therapy (ART) islatravir (ISL), the M184V and A114S resistance mutations have emerged in the human immunodeficiency virus type 1 reverse transcriptase (HIV-1 RT). These mutations markedly hypersensitize RT to the globally administered ART tenofovir disoproxil fumarate (TDF). We have solved six structures - four by X-ray crystallography and two by cryo-EM - that capture the single- and double-mutant RTs during inhibitor incorporation and...

CLASP: Language-Driven Robot Skill Selection and Composition using Task-Parameterized Learning

Announce Type: new Abstract: Enabling robots to understand and execute tasks from natural language commands while maintaining data efficiency remains challenging. Foundation models such as vision-language-action (VLA) and vision-language models (VLMs) provide intuitive interaction channels but require extensive data; task-parameterized imitation learning achieves data efficiency but lacks natural language grounding. This work bridges this gap through a modular architecture combining...

Scaling Neural Network Verification with Tensor Parallelism and Fully Sharded Data Parallelism

arXiv:2606.09377v1 Announce Type: new Abstract: Formal neural network verification -- proving that a network satisfies safety properties for \emph{all} inputs in a specified domain -- is bounded in practice by GPU memory: standard implementations of bound-propagation algorithms (IBP, CROWN, $\alpha$-CROWN) require weight and relaxation-coefficient matrices to reside entirely on one accelerator. We adapt two parallelism techniques originally developed for large-scale model training to the...