⚡️A multilingual CI linter for eliminating language barriers in global development. - HzaCode/Langlint

Large language model (LLM) queries are predominantly processed by frontier models in centralized cloud infrastructure. Rapidly growing demand strains this paradigm, and cloud providers struggle to scale infrastructure at pace. Two advances enable us to rethink this paradigm: small LMs (<=20B active parameters) now achieve competitive performance to frontier models on many tasks, and local accelerators (e.g., Apple M4 Max) run these models at interactive latencies. This raises the question: can local inference viably redistribute demand from centralized infrastructure? Answering this requires measuring whether local LMs can accurately answer real-world queries and whether they can do so efficiently enough to be practical on power-constrained devices (i.e., laptops). We propose intelligence per watt (IPW), task accuracy divided by unit of power, as a metric for assessing capability and efficiency of local inference across model-accelerator pairs. We conduct a large-scale empirical study across 20+ state-of-the-art local LMs, 8 accelerators, and a representative subset of LLM traffic: 1M real-world single-turn chat and reasoning queries. For each query, we measure accuracy, energy, latency, and power. Our analysis reveals $3$ findings. First, local LMs can accurately answer 88.7% of single-turn chat and reasoning queries with accuracy varying by domain. Second, from 2023-2025, IPW improved 5.3x and local query coverage rose from 23.2% to 71.3%. Third, local accelerators achieve at least 1.4x lower IPW than cloud accelerators running identical models, revealing significant headroom for optimization. These findings demonstrate that local inference can meaningfully redistribute demand from centralized infrastructure, with IPW serving as the critical metric for tracking this transition. We release our IPW profiling harness for systematic intelligence-per-watt benchmarking.

FOSDEM'26 HPC, Big Data, and Data Science Devroom

i have gone a little above and beyond trying to get all the features of VSCode

In January 2025, the PSF submitted a proposal to the US government National Science Foundation under the Safety, Security, and Privacy of Open Source Ecosystems program to address structural vulnerabilities in Python and PyPI. It was the PSF’s first time applying for government funding, and navigating the intensive process was a steep learning curve for our small team to climb. Seth Larson, PSF Security Developer in Residence, serving as Principal Investigator (PI) with Loren Crary, PSF Deputy Executive Director, as co-PI, led the multi-round proposal writing process as well as the months-long vetting process. We invested our time and effort because we felt the PSF’s work is a strong fit for the program and that the benefit to the community if our proposal were accepted was considerable. We were honored when, after many months of work, our proposal was recommended for funding, particularly as only 36% of new NSF grant applicants are successful on their first attempt. We became concerned, however, when we were presented with the terms and conditions we would be required to agree to if we accepted the grant. These terms included affirming the statement that we “do not, and will not during the term of this financial assistance award, operate any programs that advance or promote DEI, or discriminatory equity ideology in violation of Federal anti-discrimination laws.” This restriction would apply not only to the security work directly funded by the grant, **but to any and all activity of the PSF as a whole**. Further, violation of this term gave the NSF the right to “claw back” previously approved and transferred funds. This would create a situation where money we’d already spent could be taken back, which would be an enormous, open-ended financial risk. Diversity, equity, and inclusion are core to the PSF’s values, as committed to in our mission statement: > _The mission of the Python Software Foundation is to promote, protect, and advance the Python programming language, and to support and facilitate the growth of**a diverse and international community** of Python programmers._ Given the value of the grant to the community and the PSF, we did our utmost to get clarity on the terms and to find a way to move forward in concert with our values. We consulted our NSF contacts and reviewed decisions made by other organizations in similar circumstances, particularly The Carpentries. In the end, however, the PSF simply can’t agree to a statement that we won’t operate any programs that “advance or promote” diversity, equity, and inclusion, as it would be a betrayal of our mission and our community. We’re disappointed to have been put in the position where we had to make this decision, because we believe our proposed project would offer invaluable advances to the Python and greater open source community, protecting millions of PyPI users from attempted supply-chain attacks. The proposed project would create new tools for automated proactive review of all packages uploaded to PyPI, rather than the current process of reactive-only review. These novel tools would rely on capability analysis, designed based on a dataset of known malware. Beyond just protecting PyPI users, the outputs of this work could be transferable for all open source software package registries, such as NPM and Crates.io, improving security across multiple open source ecosystems. In addition to the security benefits, the grant funds would have made a big difference to the PSF’s budget. The PSF is a relatively small organization, operating with an annual budget of around $5 million per year, with a staff of just 14. $1.5 million over two years would have been quite a lot of money for us, and easily the largest grant we’d ever received. Ultimately, however, the value of the work and the size of the grant were not more important than practicing our values and retaining the freedom to support every part of our community. The PSF Board voted unanimously to withdraw our application. Giving up the NSF grant opportunity—along with inflation, lower sponsorship, economic pressure in the tech sector, and global/local uncertainty and conflict—means the PSF needs financial support now more than ever. We are incredibly grateful for any help you can offer. If you're already a PSF member or regular donor, you have our deep appreciation, and we urge you to share your story about why you support the PSF. Your stories make all the difference in spreading awareness about the mission and work of the PSF. How to support the PSF: * Become a Member: When you sign up as a Supporting Member of the PSF, you become a part of the PSF. You’re eligible to vote in PSF elections, using your voice to guide our future direction, and you help us sustain what we do with your annual support. * Donate: Your donation makes it possible to continue our work supporting Python and its community, year after year. * Sponsor: If your company uses Python and isn’t yet a sponsor, send them our sponsorship page or reach out to [email protected] today. The PSF is ever grateful for our sponsors, past and current, and we do everything we can to make their sponsorships beneficial and rewarding.

HIP backend patch for Numba, the NumPy aware dynamic Python compiler using LLVM. - ROCm/numba-hip

Sometimes you can trick the compiler into generating more efficient code. How can you test this optimization continues to be applied?

In 2023, SMHI was awarded development access to use the GPU partition of LUMI supercomputer. The purpose was to develop its own handwritten text recognition (HTR) models, integrated into an application known […] The post Handwritten text recognition with Dawsonia appeared first on ENCCS.

Are you interested in how artificial intelligence and supercomputing can be leveraged for the benefit of society and the environment? Join our rapidly growing group of AI experts, machine learning eng

The PeerTube instance for content genuinely made by you.