Install
The Mathematics Search Engine
Mathematics News & Resources
4Mathematics is a specialist search engine for Mathematics. Discover the latest math news and mathematical content. Part of the 4SEARCH network of topic specific search engines.
Latest Articles
VRAM usage doesn't decrease even with 4-bit quantization? What I learned from actual measurements with vLLM
2+ week, 2+ day ago (545+ words) "If you quantize a model to 4-bit, VRAM usage will be roughly 1/4"—this is a feeling you naturally develop when working with local LLMs. However, when I actually ran FP16, AWQ, and GPTQ on vLLM and measured the VRAM usage with…...
Physics education postdoc awarded prestigious fellowship for AI research
8+ min ago (325+ words) David Perl-Nussbaum, a postdoctoral research fellow in physics education research, has been awarded a prestigious fellowship by the Israel Science Foundation (ISF) to conduct research on artificial intelligence-generated computational models in physics education at CU Boulder. Perl-Nussbaum studies how higher…...
Why are local LLMs 'slow even though they fit in memory'? Understanding quantization, KV cache, and generation speed separately
1+ week, 1+ day ago (831+ words) The model loaded. There are no out-of-memory errors. Yet, you wait a long time for text to appear—with local LLMs, there is a gap between 'it runs' and 'it is comfortable to use'. When investigating the cause, it is…...
Five Classifiers, One Dataset: What I Learned About Model Choice
9+ hour, 3+ min ago (306+ words) If you hand five different classifiers the exact same data, how different do the results really look? That was the question behind my Admissions Predictions project, and the answer was more interesting than I expected. The task is binary: given…...
Hitachi boosts circuit candidate coverage up to 13.3% with LLM training
1+ day, 2+ hour ago (363+ words) The method generates options balancing competing operational goals, with potential applications in mobility, logistics and circuit design. KEY POINTS Hitachi develops LLM training technology to generate multiple options balancing conflicting performance indicators in infrastructure and industrial operations Circuit-design tests show…...
When My House Price Model Hit 100% Accuracy, I Should Have Been Worried
9+ hour, 3+ min ago (538+ words) My first script on the Kaggle House Prices data reported 100% training accuracy. I was pleased for about as long as it took to look at my feature list. The script was meant to predict SalePrice, but I had included SalePrice…...
[Latest US Findings] The Era of Prompt 'Spells' is Over. Wharton School Research Reveals 'Spec-Driven Prompting' and Practical Templates [For Copy-Paste]
1+ week, 5+ day ago (367+ words) "Take a deep breath and think step-by-step.""You are a professional marketer. Please come up with the best ideas." Are you still using these "common prompt tips found online"? In fact, studies measuring the actual effectiveness of these techniques are…...
Eight Emotions, 34,792 Tweets, and One Label I Don't Fully Trust
15+ min ago (303+ words) One example in my sentiment project still bothers me. The dataset contains the text "I have a feeling i will fail french #fuckfrench", and its label is joy. A human would read that as anxious or frustrated. I will come…...
Benchmarking LLMs: A guide to AI model evaluation
11+ hour, 48+ min ago (755+ words) KOHb - Getty Images Large language models seem to be a double-edged sword. While they can answer questions -- including questions on how to create code and test it -- the answers to those questions are not always reliable. With so many large…...
Google Metrax Brings Predefined Model Evaluation Metrics to JAX
9+ min ago (546+ words) QCon AI Boston (July 8-9, 2027): Agents. Context. Evals. Security. See how other teams make decisions. Register Now Susan Chang explains how Elastic transitioned from siloed, ad-hoc AI agent evaluations to a unified, production-grade framework. She discusses balancing LLM-as-a-judge with deterministic rules,…...