M

machine-learning

Submit, supervise and recover multi-node GPU training jobs on Slurm and Kubeflow from one CLI. Built on go-tool-base. Named for Dún Scáith, the fortress on Skye that housed Scáthach's school.

Submit and supervise multi-node training jobs behind one contract: typed failure causes, per-cluster capabilities and recovery policies as data. Scheduler adapters live in their own modules.

Self-hosted forecasting + prediction service. Five zero-shot time-series foundation models (Chronos-2, TimesFM 2.5, Moirai-2, Toto-1, Sundial) across six forecast types, plus nine supervised tabular ML backends (LightGBM, XGBoost, sklearn family) with calibrated / stacking / diversified meta-learners. Unified REST API + MCP server.

R package for Gaussian processes built from first principles: composable covariance kernels, exact GP regression with Cholesky-based numerics, analytical marginal-likelihood gradients, posterior sampling, calibration diagnostics, time-series forecasting, and heteroscedastic and sparse (FITC) extensions.

Agent skills for Ultralytics YOLO, covering models, datasets, training, tuning, inference, tracking, and export.

GitHub: https://github.com/ultralytics/skills

https://www.ultralytics.com

Official Python SDK for the Ultralytics Platform API

GitHub: https://github.com/ultralytics/sdk

https://platform.ultralytics.com/api/docs

Ultralytics YOLO27, YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking

GitHub: https://github.com/ultralytics/ultralytics

https://platform.ultralytics.com

Готовая база из 5 881 ответа по 21 тесту hh.ru - сдать с первой попытки https://it-mountain.ru/hh-tests

ein — a single vector for many things. Embedding server around EmbeddingGemma 2: many text/image/audio chunks in one POST /embed → exactly ONE weighted-centroid vector, shared 768-dim space. Node.js + TypeScript, Python sidecar, CUDA auto-detect.

Official repository of the Ruhr university Neural Network energy representation (RuNNer).

Aston University DG1AID lab repository with AI and data science notes, Python notebooks, NumPy, Pandas, search algorithms and machine learning practice.

Open industrial IoT platform for predictive maintenance: four sensor nodes, a FastAPI backend, Isolation Forest anomaly detection and a live Next.js dashboard. Reveal before failure.

Train and fine-tune open language models in Go: full training, LoRA and QLoRA on XLA, writing checkpoints and PEFT adapters the Python ecosystem loads. Named for Scáthach, who trained the heroes of the Ulster Cycle.

RuNNerASE is a collection of packages for training, evaluating, and analyzing machine learning potentials with RuNNer, the Ruhr university Neural Network energy representation.

Profile README of Diogo Ribeiro, lead data scientist, machine learning engineer and professor working on statistical machine learning, time series, causal inference and applied AI.

Evaluate language models against an OpenAI-compatible endpoint: pluggable scorers, statistics with confidence intervals, and base-versus-tuned comparison.

In-job launcher for distributed training: wraps torchrun, turns preemption signals into checkpoints, watches for hangs and GPU faults, and records why a run ended.

Kubeflow Trainer (TrainJob) adapter for go/dispatch.

Slurm adapter for go/dispatch, over SSH to a login node or through slurmrestd.

Self-learning diagnostics agent for GNU/Linux that uses unsupervised machine learning to detect anomalies in system logs in real-time.