helion kernels now ship pre-tuned on hugging face hub
meta's helion dsl and hugging face's kernels project now let developers package and distribute autotuned, portable machine learning kernels with pre-tuned configs.
machine learningsummaries filed under tools
meta's helion dsl and hugging face's kernels project now let developers package and distribute autotuned, portable machine learning kernels with pre-tuned configs.
machine learningnew research shows that safety tuning should refuse only harmful subsets of a topic, not the whole topic, to avoid over-refusal on safe prompts.
machine learninga new report links openai's ai agents to a may attack on the rubygems package repository, raising concerns about undisclosed ai-driven cyber incidents.
toolsibm's granite time series patchtst-fm-r2 offers strong zero-shot forecasting under apache 2.0 and openmdw licenses.
machine learningopenai used an unreleased model to resolve the navier-stokes millennium prize problem, sparking priority disputes with anthropic-affiliated mathematicians.
toolsa 100-step grpo run on a 350m model improves ifstruct benchmark score from 22.6% to 29.7%.
machine learningneomme is a family of 260m and 800m multilingual multimodal encoders that process text and images in a single bidirectional transformer, trained from scratch with masked discrete diffusion.
machine learningpytorch 2.14 introduces a new gpu math backend, fault-tolerant distributed training, and native linear algebra for apple silicon.
machine learninghugging face launched a library of 207 optimized webgpu kernels and a browser benchmarking tool to speed up local ai inference.
machine learninghugging face and voice arena release monsoon, a new open evaluation set for hindi and indian english speech recognition with detailed speaker metadata.
machine learninggoogle deepmind releases gemini omni 1.1 flash, a production-ready video model with scene extension, keyframe interpolation, 360p drafts, and 4k upscaling.
researchresearcher johann rehberger found a prompt injection attack that bypasses claude code's auto mode safety classifier 80% of the time.
toolsa new healing method lets a compressed, 4-bit model outperform its bfloat16 original on most benchmarks.
machine learningibm released granite 4.2, a family of dense reasoning llms in 3b, 8b, and 30b sizes, trained from scratch on 15t tokens with a multi-stage rl pipeline.
machine learninga system that spawns coding agents with pre-loaded memories from personal databases to avoid empty context windows.
researchA new GPU scheduler from Dharma-AI improves cluster utilization and priority-weighted output by planning allocations across the entire scheduling horizon instead of using FIFO order.
machine learningNew probes show leading ASR models reproduce benchmark transcripts even when audio contradicts them, inflating scores.
machine learningIBM Research shows that the right amount of agentic memory varies by model tier, with curated retrieval boosting weaker models and full guideline sets helping stronger ones.
machine learningLiquid AI published DSpark draft checkpoints for three LFM2.5 models, enabling speculative decoding that speeds up inference up to 3.2x on GPU and 2.87x on-device without changing output quality.
machine learningLiquid AI releases quantization-aware distillation checkpoints that recover 97% of BF16 accuracy for LFM2.5 models at Q4_0 memory and speed.
machine learning