913 个 · 按相关度排序
Guides LLM compression through pruning, which removes selected model weights. Covers Wanda and SparseGPT methods without retraining, accuracy comparisons, and structured weight patterns that can accelerate inference on compatible hardware.
中文简介正在准备中。当前显示英语简介。
Implement Mamba, a model architecture that processes sequences through an internal state with computation scaling linearly with input length. Covers pretrained text generation, Mamba-1 and Mamba-2 configuration, speed comparisons with Transformers, and GPU memory troubleshooting.
中文简介正在准备中。当前显示英语简介。
Choose a serving container for models deployed on SageMaker. Match the model type to a compatible image, retrieve its address from the AWS catalog, and check regional availability, CPU or GPU variants, and required host settings.
中文简介正在准备中。当前显示英语简介。
Check whether a Dynamo deployment is ready for disaggregated serving, where input processing and token generation run separately. Read-only probes inspect transport settings, GPU networking capabilities, and NIXL transfer-test readiness, then report gaps and next steps.
中文简介正在准备中。当前显示英语简介。
Add cursor-following water-ripple distortion to existing images using Shaders components. Preserve page content and layout while integrating lazy loading, accessible image fallbacks, and static display when WebGPU is unavailable or reduced motion is enabled.
中文简介正在准备中。当前显示英语简介。
Add animated backgrounds such as waves or fog with Vanta.js, which renders graphics through WebGL. Configure effects, integrate them into pages or React components, handle resizing and cleanup, and plan fallbacks for mobile devices.
中文简介正在准备中。当前显示英语简介。
Measure and improve Three.js or WebGL game performance through repeatable gameplay checks. Identify CPU or GPU bottlenecks, reduce unnecessary rendering and processing, and verify that controls, collisions, visuals, and memory remain stable.
中文简介正在准备中。当前显示英语简介。
Compare batching, request scheduling, and memory management for a single model serving instance. Assess concurrent request capacity, response times, generation speed, and GPU time per result that meets the required quality and deadline.
中文简介正在准备中。当前显示英语简介。
Estimate GPU memory, training step time, checkpoint costs, and recovery overhead for training or reinforcement learning. Compare state partitioning, recomputation, offloading, and device counts to assess whether a job can finish within its deadline.
中文简介正在准备中。当前显示英语简介。
Prepare reproducible setup and removal instructions for software environments on Open-Science SSH hosts, including Slurm clusters. Check available tools, define named activation settings, and validate the environment where jobs run after the user installs it.
中文简介正在准备中。当前显示英语简介。
Generate synthetic CT volumes with paired masks that mark anatomical regions or tumors. Configure body regions and output dimensions, then inspect image slices, label mappings, and recorded generation details for synthetic research workflows.
中文简介正在准备中。当前显示英语简介。
Fine-tune NV-Generate-CTMR MR-Brain v1 using supplied brain MRI volumes. Check the input file list, prepare training configurations, run training, and optionally generate sample volumes, producing a saved model and a structured run summary.
中文简介正在准备中。当前显示英语简介。
Plan and configure Nemotron customization workflows using existing steps for data preparation, translation, training, conversion, optimization, and evaluation. Check how outputs connect between stages and produce configuration files suited to the supplied models, data, and hardware.
中文简介正在准备中。当前显示英语简介。
Find starting configurations for training MoE models, which route work among specialized expert networks, on H100, B200, GB200, and GB300. Provides model-specific parallel layouts, communication options, and approximate performance bands to guide measured comparisons.
中文简介正在准备中。当前显示英语简介。
Choose containers and Python package sources that match a Jetson device's GPU architecture, JetPack, and CUDA environment. Get compatibility hints for PyTorch, vLLM, and related stacks before installation, using NVIDIA catalogs and Jetson AI Lab indexes.
中文简介正在准备中。当前显示英语简介。
Check whether a Dynamo deployment is ready for disaggregated serving, where input processing and token generation run separately. Read-only probes inspect transport settings, GPU networking capabilities, and NIXL transfer-test readiness, then report gaps and next steps.
中文简介正在准备中。当前显示英语简介。
Fine-tune GR00T or openpi robot policies using compatible LeRobot data. Check dataset requirements, preview the training configuration, run training, and verify that a saved checkpoint, a file containing the trained model, loads through the inference backend.
中文简介正在准备中。当前显示英语简介。
Install cuOpt for Python, C, or use as a REST server, a service called over HTTP. Choose pip, conda, or Docker to match the environment, verify installation, and troubleshoot package or CUDA mismatches.
中文简介正在准备中。当前显示英语简介。
Pretrain a new KERMT chemistry model from a molecular corpus, starting with random weights. The skill validates the data, builds fresh vocabularies, estimates runtime, and launches containerized training with logs and records for reproducing the run.
中文简介正在准备中。当前显示英语简介。
Run Python code on cloud GPUs using Modal serverless platform. Use when you need A100/T4/A10G GPU access for training ML models. Covers Modal app setup, GPU selection, data downloading inside functions, and result handling.
中文简介正在准备中。当前显示原文说明。
Architecture rules for GPU rendering with TypeGPU/WebGPU. Use when adding or changing render passes, shaders, frame orchestration, GPU resources, culling, terrain, shadows, or visual performance work.
中文简介正在准备中。当前显示原文说明。
Set up and manage NVIDIA GPU servers for AI workloads — driver installation, CUDA toolkit, container toolkit, MIG partitioning, GPU health monitoring, and multi-GPU configuration for LLM inference and training.
中文简介正在准备中。当前显示原文说明。
Implement or debug Hugging Face Spaces ZeroGPU code using `@spaces.GPU`, including isolation, state, duration, sizing, and CUDA-wheel constraints. Use for ZeroGPU-specific behavior; use huggingface-spaces for general hosting.
中文简介正在准备中。当前显示原文说明。
The investigation harness for "why is this slow?" over a Pulp Perfetto trace (.pftrace). Runs the hypothesis→query→drill-down chain-of-evidence loop autonomously and returns a plain-English root cause + evidence + a concrete fix. TRIGGER on "why is my plugin slow to open", "find the slowest frames", "why is the UI stuttering", "why is my plugin using so much CPU", "which DSP node is expensive", "the load meter looks calm but CPU is pinned", or a named `pulp trace gpu-startup|gpu-health|gpu-probe` analysis. Ships Pulp-specific hints for dsp, frame, js, gpu, and cross-platform symptoms.
中文简介正在准备中。当前显示原文说明。