Loading...
Loading...
Found 423 Skills
Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.
Expert guide for WebGL API development including 3D graphics, shaders (GLSL), rendering pipeline, textures, buffers, performance optimization, and canvas rendering. Use when working with WebGL, 3D graphics, canvas rendering, shaders, GPU programming, or when user mentions WebGL, OpenGL ES, GLSL, vertex shaders, fragment shaders, texture mapping, or 3D web graphics.
Trains and fine-tunes vision models for object detection (D-FINE, RT-DETR v2, DETR, YOLOS), image classification (timm models — MobileNetV3, MobileViT, ResNet, ViT/DINOv3 — plus any Transformers classifier), and SAM/SAM2 segmentation using Hugging Face Transformers on Hugging Face Jobs cloud GPUs. Covers COCO-format dataset preparation, Albumentations augmentation, mAP/mAR evaluation, accuracy metrics, SAM segmentation with bbox/point prompts, DiceCE loss, hardware selection, cost estimation, Trackio monitoring, and Hub persistence. Use when users mention training object detection, image classification, SAM, SAM2, segmentation, image matting, DETR, D-FINE, RT-DETR, ViT, timm, MobileNet, ResNet, bounding box models, or fine-tuning vision models on Hugging Face Jobs.
Route Unity and Unreal frame-time complaints into one bottleneck-first profiling brief. Use when the main job is interpreting profiler screenshots, `stat unit` / `stat gpu` output, benchmark-route complaints, or Steam Deck / target-device review packets; choosing the smallest useful next capture; naming one primary bottleneck family; and deciding whether to stay with quick packets, move to an engine-native profiler, or escalate further. Route generic app/service tuning to `performance-optimization`, build/editor/package failures to `game-build-log-triage`, broader game-production coordination to `bmad-gds`, and mixed demo/community feedback to `game-demo-feedback-triage`.
Profiles and optimizes TensorRT-LLM host/CPU overhead using line_profiler (with nsys support planned). Runs iterative profile-analyze-optimize-validate rounds. Use when GPU utilization is low or optimizing PyExecutor throughput.
Search for products on Newegg (https://www.newegg.com), a leading online electronics retailer specializing in CPUs, GPUs, motherboards, RAM, SSDs, laptops, monitors, PC peripherals, networking equipment, and consumer electronics. Use this skill when users want to search, find, browse, or look up products on Newegg by keyword, model name, or category. Supports price range filtering, pagination, and sort options. Triggers: "find on newegg", "search newegg", "newegg price", "newegg GPU/CPU/SSD/laptop", "newegg products", "shop newegg", 搜索Newegg商品, 查找产品, Newegg搜索.
Refactor module-level imports of heavy native dependencies (torch, tensorflow, jax) to function-internal lazy imports. Use when a library import triggers SIGSEGV or slow startup in environments where the native dep cannot fully initialize (e.g., CUDA driver absent, triton C-extension segfault, missing GPU). Predicts test-side effects from removed module attributes.
Start a new Blockbench model or scene from a short intake form and deliver a render-ready Blender scene. Asks for quality, texture style (PBR 2K/HD/standard, vanilla Minecraft, classic Blockbench), animation (keyframed or Havok physics), scene type (scene, vignette, isometric room), subject, output folder and render preset; builds .bbmodel files with the headless Blockbench MCP server and parallel agents; imports them into Blender; and writes a render.cmd that renders stills on every GPU and assembles them into a video with FFmpeg. Use when the user runs /blockbench-new-model or asks for a new Blockbench model delivered as a rendered Blender scene or video.
Complete design system reference for full-stack applications. Covers colors, typography, spacing, component patterns (shadcn/ui), effects, GPU-accelerated animations, and WCAG AA accessibility. Use when implementing UI, choosing colors, applying spacing, creating components, or ensuring brand consistency.
Cloud GPU processing via RunPod serverless. Use when setting up RunPod endpoints, deploying Docker images, managing GPU resources, troubleshooting endpoint issues, or understanding costs. Covers all 5 toolkit images (qwen-edit, realesrgan, propainter, sadtalker, qwen3-tts).
This skill should be used when users want to train or fine-tune language models using TRL (Transformer Reinforcement Learning) on Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment. Includes guidance on the TRL Jobs package, UV scripts with PEP 723 format, dataset preparation and validation, hardware selection, cost estimation, Trackio monitoring, Hub authentication, and model persistence. Should be invoked for tasks involving cloud GPU training, GGUF conversion, or when users mention training on Hugging Face Jobs without local GPU setup.
Deploys remote development environments with SSH access on TrueFoundry. Use when setting up VS Code Remote workspaces, cloud development machines, GPU dev boxes, or remote coding environments with SSH access.