Tootfinder

DECA: A Near-Core LLM Decompression Accelerator Supporting Out-of-Order Invocation
Gerasimos Gerogiannis (Intel Corporation,University of Illinois at Urbana-Champaign), Stijn Eyerman (Intel Corporation), Evangelos Georganas (Intel Labs), Wim Heirman (Intel Corporation), Josep Torrellas (University of Illinois at Urbana-Champaign)
https://

DECA: A Near-Core LLM Decompression Accelerator Supporting Out-of-Order Invocation
To alleviate the memory bandwidth bottleneck in Large Language Model (LLM) inference workloads, weight matrices are stored in memory in quantized and sparsified formats. Hence, before tiles of these matrices can be processed by in-core generalized matrix multiplication (GeMM) hardware engines, they need to be dequantized and de-sparsified. This is currently performed in software with vector operations. Unfortunately, this approach delivers only modest performance. Moreover, it is hard to unders…

MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents
Zijian Zhou, Ao Qu, Zhaoxuan Wu, Sunghwan Kim, Alok Prakash, Daniela Rus, Jinhua Zhao, Bryan Kian Hsiang Low, Paul Pu Liang
https://arxiv.org/abs/2506.15841

MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents
Modern language agents must operate over long-horizon, multi-turn interactions, where they retrieve external information, adapt to observations, and answer interdependent queries. Yet, most LLM systems rely on full-context prompting, appending all past turns regardless of their relevance. This leads to unbounded memory growth, increased computational costs, and degraded reasoning performance on out-of-distribution input lengths. We introduce MEM1, an end-to-end reinforcement learning framework …

Performance of Confidential Computing GPUs
Antonio Mart\'inez Ibarra, Julian James Stephen, Aurora Gonz\'alez Vidal, K. R. Jayaram, Antonio Fernando Skarmeta G\'omez
https://arxiv.org/abs/2505.16501

Performance of Confidential Computing GPUs
This work examines latency, throughput, and other metrics when performing inference on confidential GPUs. We explore different traffic patterns and scheduling strategies using a single Virtual Machine with one NVIDIA H100 GPU, to perform relaxed batch inferences on multiple Large Language Models (LLMs), operating under the constraint of swapping models in and out of memory, which necessitates efficient control. The experiments simulate diverse real-world scenarios by varying parameters such as …

I just found out by trying, that on my Linux install
Alt Gr - = – (why would anyone use em dashes? those are way too long anyway. en dashes is where it's at.)
Alt Gr . = … (OMG, yes! Windows with its Alt 0,1,3,3 can go suck it!)
Alt Gr , = · (… tf is a middledot?)
Thanks for the inspiration, Wolfgang!
htt…

PUL: Pre-load in Software for Caches Wouldn't Always Play Along
Arthur Bernhardt, Sajjad Tamimi, Florian Stock, Andreas Koch, Ilia Petrov
https://arxiv.org/abs/2506.16976

PUL: Pre-load in Software for Caches Wouldn't Always Play Along
Memory latencies and bandwidth are major factors, limiting system performance and scalability. Modern CPUs aim at hiding latencies by employing large caches, out-of-order execution, or complex hardware prefetchers. However, software-based prefetching exhibits higher efficiency, improving with newer CPU generations. In this paper we investigate software-based, post-Moore systems that offload operations to intelligent memories. We show that software-based prefetching has even higher potential i…

This https://arxiv.org/abs/2411.04215 has been replaced.
initial toot: https://mastoxiv.page/@arXiv_qu…

Sound certification of memory-bounded quantum computers
The rapid advancement of quantum hardware calls for the development of reliable methods to certify its correct functioning. However, existing certification tests often fall short: they either rely on flawless state preparation and measurement or lack soundness guarantees, meaning that they do not rule out incorrect implementations of the target operations by a quantum device. We introduce an approach, which we call quantum system quizzing, for the certification of quantum gates in a practical s…

#Linux tip for people running Rocky Linux #VFX workstations (or Alma or RHEL I guess) who are looking for a solution to the system locking up once you're running out of memory: install systemd-oomd - the out of memory killer daemon. It's not enabled by default but I think it is on distributions that…

Thanks @… for your article on Docker cgroup on Raspberry Pi, helped me out of a pickle!
https://akashrajpurohit.com/…

Resolving Missing Memory Stats in Docker Stats on Raspberry Pi
Running Docker containers on your Raspberry Pi can be a great way to explore different software and services. However, encountering issues with retrieving memory usage information using docker stats can be frustrating. In this blog post will dive into the common reasons behind missing memory stats and guide you through the troubleshooting steps to resolve them.

I’ve never seen this memory issue on #MacOS before. I had left Safari with Mela’s share sheet/importer app snippet open overnight and came back to a nearly unresponsive computer. I assume one of these has a memory leak. I was able to reboot via the shell using SSH from another device.

Hey everyone!
Resources 1.8 has landed with new features such as showing the link type and speed of various devices and better support for more hardware like the Raspberry Pi's GPU.
And, of course, it features the latest GNOME 48 runtime!
Feel free to check it out! :)
https://flathub.org/apps…

Maybe some Linux Kernel expert could help me with this:
Due to the nature of my work, I regularily run a very memory intensive process and my system runs out of memory during very few peak moments.
The system can't swap out memory fast enough and freezes. If I disable swap, the process gets killed instead.
Is there any way I can keep the system responsive? Ideally, only the offending process would get slowed down, but not other applications.

Hey everyone!
Resources 1.8 has landed with new features such as showing the link type and speed of various devices and better support for more hardware like the Raspberry Pi's GPU.
And, of course, it features the latest GNOME 48 runtime!
Feel free to check it out! :)
https://flathub.org/apps…

The liberal obsession with optics serves the right and persuades no one. There is literally an active ethnic cleansing happening in the US right now, and the only thing that matters is making that as hard as possible to carry out.
Anarchists destroying intelligence assets saves lives. Every escooter thrown at a cop car is one less escort for a goon too afraid to kidnap random brown people without being flanked by a branch full of bad apples. Spray paint is not violence. Vandalism is not violence. Community self defense in all forms is legitimate.
Make no mistake, these raids are about changing demographics. Demographic trends have been shifting blue for a long time, and the right has, for a long time, been blaming "white replacement." Conspiracy theory aside, Democrats have also been relying on the growth of black and brown voters as a block. The nuances of whiteness as an identity are lost on the current administration and their supporters. They see that "white people will be a minority by 2050" and equate that with the "end of Western Civilization."
The only way to "save Western Civilization" is to change those demographics. Forced birth and forced removal are two sides of the same white nationalist objective. Of course they can't have due process, because they need to be able to kidnap anyone who they see as a threat to their demographic future.
They don't care about optics. The plan is to murder away any threat and flood everyone else with propaganda. There is no mythical middle. There's no one unconvinced. They know this, but they win when democrats buy that myth and save the police the work of policing the protests.
If your protest is 90% "peaceful," they'll take pictures of the 10% that isn't. If it's 99% peaceful, they'll shoot rubber bullets and teargas until someone throws a brick and take 100 pictures from a dozen angles. If its 100% "peaceful" and no one can be provoked, they'll generate pictures with AI or photoshop like they did during the George Floyd uprising and the pictures from the CHOP/CHAZ. Do you have literally no memory?
#USPol #FiftyFiftyOne #50501movenent #resistance #NoKingsDay #NoKingsDayOfAction

Any greybeards out there have a memory of magnetic card data storage media "for word processing systems?" New to me!

This https://arxiv.org/abs/2504.09432 has been replaced.
initial toot: https://mastoxiv.page/@a…

Nanoscale Quantum Imaging of Spin Dynamics using a Hybrid 2D/3D System
Spin defects in solids offer promising platforms for quantum sensing and memory due to their long coherence times and compatibility with quantum networks. Here, we integrate a single nitrogen-vacancy (NV) center in diamond with scanning probe microscopy to discover, read out, and spatially map arbitrary spin-based quantum sensors at the nanoscale. Using the boron vacancy (V$_\mathrm{B}^-$) center in hexagonal boron nitride$\unicode{x2013}$an emerging two-dimensional spin system$\unicode{x2013}$…

Read "What went wrong with capitalism" by Ruchir Sharma.
A mildly interesting description of the major events in world an US economics in the last 50 years. Might
be a fair summary for anyone who didn't live through it or has a poor memory.
In short he thinks what went wrong was government bailing out failure leading to massive debts and increa
sed inequality.
Governments took over all the things instead of letting capitalism sort them out, he reckons, and wheneve
r a big industry or company fails you just get socialism for the rich and a bail-out from new printed money.
Easy cheap money, constant bail-outs, government intervention, leading to zombie companies racking up every larger debt to exist, billionaires who can't fail due to government support, and a stock market that's up-only bringing a flood of inefficiently-allocated capital.
Is he right? I mean, maybe, sort of. But when an industry really can't be allowed to fail, say water supply and waterway management, allowing private capital to extract maximum resources from it isn't the best method to manage it in the first place. No wonder they need bail-outs. Capitalism fails here because capitalism isn't the right solution here. We need publicly owned national services, not robber barons without
bailouts.
So, you know, half right. Perhaps these are some of the reasons why capitalism fails, but also we shouldn't even be trying to apply capitalism to every single thing in the first place.
#reading #capitalism #economics #RuchirSharma