Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
DiffusionGemma, a text diffusion model, achieves a 2.2× step-speedup and 31-point GPU compute efficiency gain over autoregressive decoding, according to a performance review by an ML engineer. The comparison shows diffus…
A systematic study of 60 language model benchmarks found that nearly half exhibit saturation, with rates increasing with age, and that resilience to saturation is impacted by expert-curation, not by public test data. The…
A production-focused playbook from the WildOctopus team outlines when to use transfer learning, full fine-tuning, LoRA, QLoRA, continued pretraining, and knowledge distillation for large language models, recommending LoR…
Flyte 2, the durable AI runtime from the Linux Foundation, is now generally available, offering a ground-up rebuild that is faster, more scalable, and supports durable distributed AI workflows using regular Python. The n…
Cursor, the AI coding company, open-sourced Mixture-of-Kittens (MoK), a deterministic MoE training megakernel for NVL72s that fuses communication and computation into a single kernel, achieving a 1.41x increase in end-to…
Cisco IT has operationalized its AI-driven voice security by integrating Splunk Cloud Platform to ingest, correlate, and visualize millions of Call Detail Records (CDRs) from Cisco Unified Communications Manager (CUCM), …
Cisco IT reduced toll fraud by 70% and manual investigation effort by 60% by applying AI and machine learning to its enterprise voice security, using a composite risk-scoring engine that combines rules with ML across its…
A synthesis paper by Yiting Lu et al. published in MDPI's Veterinary Sciences journal describes how deep learning models are being applied to monitor and manage viral diseases in livestock, offering a way to detect outbr…
Independent researcher Alpamys Makazhan released Soup, a Show HN project that fine-tunes a full Llama-3.1-8B model in NF4 quantization with a 3.32 GB VRAM peak at 119.6 tokens/sec on a 4 GB RTX 3050 Laptop GPU under Wind…
In a benchmark run on Thursday, Fable and Sol tied at 33.3% (10/30 strict checkpoint passes) on SlopCodeBench, with Expo and Kimi K3 following at 26.7% and 23.3%, respectively. The evaluation, conducted by HumanLayer acr…
Oleksii Vasylenko released g6k-rs, a research-oriented Rust library for lattice reduction, enumeration, and sieving on CPUs, Apple Metal, and NVIDIA CUDA, available on GitHub. The library includes floating-point and arbi…
NVIDIA released Alpamayo 2 Super, an open 34-billion-parameter reasoning vision-language-action model for autonomous vehicle development, combining a 32-billion-parameter Cosmos 3 Super Reasoner with a 2-billion-paramete…
Roboflow published a guide comparing zero-shot and fine-tuned computer vision models, recommending a hybrid approach where zero-shot models prototype and generate training data before fine-tuning for production. The guid…
AI model distillation, a process where smaller 'student' models learn from larger 'teacher' models like ChatGPT, Claude, and Gemini, is becoming a divisive issue as the U.S.-China AI race intensifies. The technique skips…
A new guide on measuring transformer inference performance details metrics such as latency, time to first token (TTFT), time per output token (TPOT), throughput, memory usage, utilization, and cost per token, emphasizing…
Researchers from Microsoft Research and Paige, now part of Tempus, developed PRISM2, a pathology foundation model trained on tissue images and language from real pathology reports, and reported in Nature Medicine that it…
A single AMD MI300X GPU with 192 GB of HBM3 now serves DeepSeek's 284B-parameter DeepSeek-V4-Flash-0731 checkpoint in mixed FP4+FP8 format, requiring nine patch overlays against a vLLM ROCm nightly plus custom AITER tuni…
A study submitted to arXiv on 14 Jun 2024 and revised on 10 Apr 2025 shows that physical networks of nodes connected by resistive edges can learn diverse functions by adapting edge conductances to lower a cost function, …
Arm Holdings is advancing its neural graphics portfolio, including Neural Super Sampling (NSS), Neural Frame Rate Upscaling (NFRU), and Neural Super Sampling and Denoising (NSSD), to enhance mobile gaming visuals while m…
Fine-tuning large language models can cause catastrophic forgetting, where new knowledge overwrites old knowledge, making models specialized but less versatile. Techniques like LoRA and multi-task training can mitigate t…