Skill UI

A comprehensive guide for developers building AI demos and GPU compute tasks on Hugging Face Spaces using ZeroGPU hardware. It covers best practices for writing and reviewing code that utilizes the `@spaces.GPU` decorator, managing runtime duration and quota, and understanding the specific CUDA availability model and process isolation rules required for reliable ML deployment.

A comprehensive guide detailing the entire model lifecycle for RuView. It supports multiple advanced tracking methods including camera-free pose estimation (WiFlow), camera-supervised methods using MediaPipe, robust contrastive embeddings (RuVector), and domain generalization (MERIDIAN). Covers training, evaluation, and deployment across local GPU and cloud platforms.

Provides comprehensive tools for managing high-performance GPU compute resources, including querying live inventory (e.g., H100/H200), submitting authenticated fixed-rate Requests for Quote (RFQ), inspecting procurement status, and validating nodes. Use this when planning to rent or purchase compute capacity and require detailed boundary explanations. Includes device credential management and node qualification.

A comprehensive guide for positioning research manuscripts for the ACM SIGGRAPH I3D Symposium. It covers core topics like real-time rendering, interactive 3D systems, and GPU algorithms, advising authors on how to frame their work, structure evidence, and compare it against related academic venues to maximize submission success.

A comprehensive guide for authors targeting the ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP). This skill helps assess if a computer science manuscript fits the venue, focusing on techniques like parallel programming models, compilers, runtimes, and multicore/GPU systems. It provides structured advice on framing the contribution, identifying evidence gaps, and positioning the work against related top-tier conferences.

Predict regulatory features, gene structure, and expression from DNA sequences using hosted transformer DNA language models via REST API or MCP server. Six tasks: promoter, splice, enhancer, chromatin, expression, annotation. No local GPU needed.

A comprehensive project management guide for the MLSys conference submission cycle. It covers the entire research lifecycle, from initial venue selection and track choice through paper submission, review responses, artifact evaluation, and final conference presentation. The guide emphasizes backward planning, managing critical resources like shared GPU time, and mitigating common failure modes.

A comprehensive guide for researchers detailing how to structure the 'Related Work' section for top-tier parallel computing conferences like PPoPP. It emphasizes 'delta-first' comparison, covering specific technical lanes (concurrent data structures, runtimes, GPU acceleration, memory models), and ensuring proper academic positioning against major parallel systems venues.

This guide outlines the necessary steps to ensure that parallel computing performance papers are scientifically reproducible. Since parallel results depend heavily on machine architecture, topology, and runtime conditions, proper documentation must cover everything from CPU/GPU specifications and interconnects to thread pinning policies, random seeds, and compiler flags. The focus is on proving the scaling trend, not just the absolute numbers, allowing results to hold true across different hardware platforms.

This skill manages the loading and efficient caching of OpenMed clinical/biomedical Named Entity Recognition (NER) models. Models are downloaded once from the Hugging Face Hub, enabling full offline operation afterward. It supports model reuse across multiple calls, configuring cache directories, specifying devices (CPU/GPU), and handling various model identifiers (registry keys, full HF IDs, or local paths).

This skill guides users on deploying OpenMed models completely on-device for maximum privacy and edge computing. It details running workflows using three backends: MLX (Apple Silicon), CoreML (iOS/macOS), and ONNX/WebGPU (cross-platform/browser). The process involves converting, optionally quantizing, and running models locally, ensuring PHI never leaves the host device.

Apply Istio/Linkerd service mesh patterns to AI inference traffic: strict mTLS, per-service authorization, egress allowlists, canary and A/B model routing, circuit breaking, and cross-cluster resilience for GPU-backed model services.