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.

Probabilistic deep-learning framework built on PyTorch for single-cell omics; handles batch correction, integration, multimodal and spatial data, differential expression, and transfer learning with unified APIs.

A comprehensive skill for ingesting, analyzing, and manipulating multi-modal video and audio content. Supports real-time desktop capture, live stream monitoring (RTSP), and file uploads (URL/local). Capabilities include advanced indexing (visual, semantic, temporal), precise moment searching, professional timeline editing (subtitles, overlays, dubbing), transcoding, and generating actionable alerts. Ideal for content creators, monitoring systems, and complex media workflows.

Provides a detailed, multi-platform workflow to monitor long-running computational experiments. It checks the status of jobs across various environments (SSH, vast.ai, Modal), collects screen logs, fetches structured JSON results, and pulls detailed training metrics (loss curves, accuracy) from Weights & Biases (W&B). Essential for MLOps and automated result review.

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.

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.

Build or audit reproducible research workflows with NeuroKit2 for physiological time-series: preprocessing, event/interval analysis, multimodal alignment, variability and complexity on ECG, RSP and EDA signals.

A comprehensive skill for ingesting, analyzing, and manipulating multi-modal video and audio content. Supports real-time desktop capture, live stream monitoring (RTSP), and file uploads (URL/local). Capabilities include advanced indexing (visual, semantic, temporal), precise moment searching, professional timeline editing (subtitles, overlays, dubbing), transcoding, and generating actionable alerts. Ideal for content creators, monitoring systems, and complex media workflows.

A comprehensive strategic guide for authors targeting the ACM International Conference on Multimedia Retrieval (ICMR). This skill helps authors assess the fit of their multimedia retrieval, cross-modal search, or representation learning work for the venue. It provides specific advice on framing the contribution, defining necessary evidence (e.g., ranking metrics, baselines), and structuring the manuscript to meet ICMR's high standards.