Skill UI

NeMo Curator streams GPU-accelerated curation for LLM training, with super-fast fuzzy deduplication, multimodal quality filters, semantic dedupl, PII redaction, NSFW checks, and RAPIDS-scaled pipelines for prepping massive corpora.

Ray Data offers scalable distributed data processing for ML workloads, streaming across CPU/GPU clusters and integrating with PyTorch, TensorFlow, and Ray Train. Use it for batch inference, ETL, and multi-modal preprocessing from a laptop up to hundreds of nodes.

Ray Train orchestrates distributed PyTorch, TensorFlow, and HuggingFace workloads from laptops to clusters, automating GPU placement, fault tolerance, checkpointing, and hyperparameter sweeps so teams can scale massive model training with minimal code changes.

This advanced workflow orchestrates complex, multi-node GPU clusters on Vast.ai. It is designed for large-scale distributed machine learning training, automatically handling spot instance preemption and resuming jobs from checkpoints. Furthermore, it provides comprehensive cost analysis by tracking billing history, allowing users to optimize GPU spending and manage cluster teardown efficiently.

A comprehensive guide to optimizing deep learning workflows and maximizing efficiency on cloud GPU platforms like Vast.ai. Learn techniques for selecting optimal instances by performance-per-dollar, reducing startup latency, tuning data pipelines (DataLoader, pin_memory), and scaling multi-GPU training using PyTorch and Bash scripting.

PyGraphistry is a Python interface to the Graphistry platform for GPU-accelerated graph visualization, letting analysts render millions of nodes/edges in interactive dashboards to surface anomalies and relationships quickly.

This comprehensive tool manages the complete lifecycle of ML experiments, providing robust deployment capabilities across diverse environments. It supports running training jobs on local GPUs, dedicated remote servers, Vast.ai, and serverless platforms like Modal. The workflow handles environment detection, code synchronization (rsync/git), dependency pinning, and integrates advanced logging tools (e.g., W&B), ensuring reliable, scalable, and reproducible ML training runs.

This skill automates the process of renting, managing, and terminating cloud GPU instances on vast.ai. It analyzes your AI/ML task—determining VRAM, required GPU count, and estimated runtime—and then searches for the most cost-effective and reliable GPU offers, presenting optimized options for your training workflow.

Collection of battle-tested PyTorch training recipes covering architecture selection, optimizers, LR scheduling, precision tips, and debugging to train or fine-tune LLMs, vision, diffusion, and biomedical models with high GPU throughput.

Utilize Modal, a serverless GPU cloud, for running complex machine learning workloads, including training, fine-tuning, and inference. This tool provides a zero-config environment, eliminating the need for SSH setup or Docker management. It features automatic scale-to-zero billing and local-first execution, making high-powered GPU compute accessible for debugging and small-scale experiments.

Utilize Google's TimesFM foundation model for zero-shot univariate time series forecasting without training custom models. Supports sales, sensor, and weather data with point forecasts and prediction intervals. Includes mandatory system checks for RAM and GPU before inference to prevent crashes.

This guide details how to set up and execute large-scale, distributed GPU training jobs on the CoreWeave platform. It covers both single-node multi-GPU setups and multi-node training utilizing PyTorch DDP and Kubernetes resources (A100/H100). This workflow is ideal for fine-tuning large language models (LLMs) or running complex deep learning models that require high-performance, shared-storage compute clusters.