“nemo”的搜索结果

Plan long-sequence training for MoE models, which divide work among expert networks, in Megatron Bridge. Use experiment-based guidance to balance parallelism, selective recalculation, and routing choices when investigating memory shortages or reduced training throughput.

中文简介正在准备中。当前显示英语简介。

• Planning long-sequence MoE training configurations
• Investigating memory shortages during long-context training
• Reviewing throughput drops after a code change

Tune Megatron Bridge training for MoE models, which route tokens to specialized experts, by overlapping expert communication with computation. Check dispatcher requirements, avoid conflicting settings, and compare steady training performance while changing one overlap setting at a time.

中文简介正在准备中。当前显示英语简介。

• Reduce exposed token dispatch and combine time
• Check requirements for expert communication overlap
• Configure flex dispatch and resolve overlap conflicts

Configure and validate sequence packing, which combines training examples into token sequences, and long-context training in Megatron-Bridge. Compare offline, in-batch, and Energon online approaches while checking batch sizes, parallelism requirements, and model compatibility.

中文简介正在准备中。当前显示英语简介。

• Configure offline packing for text model fine-tuning
• Choose packing methods for vision-language models
• Check parallelism constraints for long-context training

Convert single-node training or inference commands into multi-node Slurm batch jobs. Choose an appropriate process launch method, configure shared storage and containers, and investigate communication timeouts, launch errors, and GPU memory shortages.

中文简介正在准备中。当前显示英语简介。

• Converting a single-node command into a multi-node sbatch job
• Investigating NCCL communication timeouts
• Estimating GPU memory needs for MoE models

Helps tune activation recomputation in Megatron Bridge, recalculating intermediate training values to reduce GPU memory use. Guides selective or full-layer settings, checks compatibility, and compares memory peaks with training speed to assess whether a configuration completes training steps.

中文简介正在准备中。当前显示英语简介。

• Investigate training failures caused by activation memory
• Choose recomputation targets for attention or MoE models
• Compare GPU memory savings against training speed

Plan and tune training for vision-language models that use MoE, a mixture of expert networks, in Megatron Bridge. Compares FSDP and 3D parallel training, with guidance on memory use, batch sizes, and performance checks using realistic image inputs.

中文简介正在准备中。当前显示英语简介。

• Choosing a parallel training approach for a new model
• Investigating training failures and out-of-memory errors
• Comparing performance with realistic image inputs

Set up Earth2Studio with uv or pip using commands matched to your environment and planned models. The skill guides optional dependency selection, cache configuration, installation checks, and dependency troubleshooting while leaving package installation to the user.

中文简介正在准备中。当前显示英语简介。

• Install Earth2Studio in a new or existing Python project
• Identify optional packages for planned models
• Configure cache locations and download timeouts

Prepare an environment for evaluating speech recognition on clinical terms. Checks the API key and Python dependencies, then tests NVIDIA's hosted services by generating speech from one synthetic sentence and transcribing it before the next stage.

中文简介正在准备中。当前显示英语简介。

• Preparing a clinical speech recognition evaluation
• Checking API access from a fresh environment
• Testing speech synthesis and transcription together

Predict the structure of a single protein chain from its amino acid sequence using OpenFold2 NIM. Run through NVIDIA's hosted API or local Docker, optionally add sequence alignments and structural templates, and save predictions with confidence information.

中文简介正在准备中。当前显示英语简介。

• Predicting a single protein chain's structure from its sequence
• Adding A3M sequence alignments or mmCIF structural templates
• Comparing predictions from multiple model parameter sets

FAIR-aligned exposure quantification: turns a pile of recon findings into a defensible 0-100 + A-F org risk score (Likelihood x Impact, three ownership-aware factors: exposure/threat/impact), an ownership + proof demotion cap so unproven or weakly-owned findings can't inflate the number, a $-denominated FAIR loss-magnitude estimate (IBM/Ponemon per-record cost bands, cross-source record dedup, threat-factor annualization), attack-path amplification (curated red-team chain catalog + generic graph-walk engine, with a kill-chain vs shared-fate honesty gate), and a board-ready one-pager deliverable (hero $ + letter grade + top-3 findings + top attack path + the ask). Extends osint-methodology's severity rubric and client deliverable templates with quantification. Passive analysis only -- operates on findings already collected, no target traffic, no API keys. Use when asked to score risk, quantify exposure, estimate breach cost, build a board report, translate technical findings to dollars, or explain why a grade or dollar figure came out the way it did.

中文简介正在准备中。当前显示原文说明。

Create study sessions using recall questions, chunking, mnemonics and explicit review intervals. Use for this Bot Generator Bot template; creates drafts or simulations within its declared scope.

中文简介正在准备中。当前显示原文说明。

Designs memory palace structures with spatial layouts and domain organization. Use when creating a new palace or planning knowledge architecture by hand.

中文简介正在准备中。当前显示原文说明。

CesiumJS viewer setup - Viewer, CesiumWidget, widgets, Ion token, Scene configuration, SceneMode, factory helpers, geocoders, platform services. Use when initializing a CesiumJS application, configuring viewer widgets, setting Ion access tokens, creating default terrain or imagery, or bootstrapping a 3D globe.

中文简介正在准备中。当前显示原文说明。

把任意主题做成讲解/科普视频并渲染成 MP4（HTML/CSS/GSAP 画面 + 确定性逐帧渲染 + 配音 + 词级硬字幕）。触发场景：要做科普/讲解/教学/知识/产品类视频、「讲一下 X 做成视频」、要挑某种视觉风格（极简/数据/赛博/电影感/品牌）出片、要出抖音封面/竖版封面/9:16 封面、要用 html-video 那种模板化画面但更稳的音画同步、anything2explainer 换 HTML 渲染，或提到 html-explainer / HTML 讲解视频 / explainer video / MG 视频。内置 23 个模板风格（8 类）与 40+ 动作词汇的动效库，主题驱动模板编排；支持 4K60、快门运动模糊、多进程提速、断点续渲；流程含六个用户确认点（渲染通道是硬闸门）。独立可移植，不依赖 html-video / anything2explainer。

中文简介正在准备中。当前显示原文说明。

Implement comprehensive safety guardrails for LLM applications including content moderation (OpenAI Moderation API), jailbreak prevention, prompt injection defense, PII detection, topic guardrails, and output validation. Essential for production AI applications handling user-generated content. Use when ", guardrails, content-moderation, prompt-injection, jailbreak-prevention, pii-detection, nemo-guardrails, openai-moderation, llama-guard, safety" mentioned.

中文简介正在准备中。当前显示原文说明。

Expert guide for AI Safety, Governance, and Responsible AI in production — Constitutional AI enforcement, runtime guardrails (NeMo Guardrails 2.0, Llama Guard 3), bias auditing, hallucination detection, EU AI Act compliance, model cards, and content provenance / Panduan ahli untuk Keamanan AI, Tata Kelola, dan AI Bertanggung Jawab di produksi.

中文简介正在准备中。当前显示原文说明。

Vue 3 SFC structure, Composition API + <script setup>, file naming, project layout, props/emits/slots typing, defineModel, composables, lifecycle, watchers, UI library detection. Vue 2 fallback notes. Use this skill to: - Structure a new SFC with `<script setup>`, template, scoped styles. - Type props/emits/slots correctly via macros. - Pick `ref` vs `reactive` (prefer `ref`). - Build composables that compose cleanly. - Apply lifecycle hooks correctly. - Detect UI library (Vuetify/Quasar/PrimeVue/Naive UI/Element Plus/shadcn-vue) and mirror its patterns. Do NOT use this skill for: - State management lib choice (see vue-state-management). - Routing (see vue-routing). - Forms (see vue-forms). - Testing (see vue-testing).

中文简介正在准备中。当前显示原文说明。

Contribute to OpenClaw ecosystem projects (OpenClaw, NemoClaw, NanoClaw) through a structured 9-step workflow: target verification, codebase exploration, parallel audit, finding cross-reference, and pull request creation. Emphasizes false positive prevention and project convention adherence.

中文简介正在准备中。当前显示原文说明。

Local speech-to-text on Apple Silicon with two cross-checking MLX engines: NVIDIA Parakeet TDT v2 (parakeet-mlx) and Qwen3-ASR 1.7B (mlx-audio, hotword biasing). Batch JSONL with word timestamps, confidence, provenance, and an agreement verdict. No NeMo or PyTorch. Use when transcribing audio files, identifying game or legacy voice clips by content, telling speech from SFX, serving an OpenAI-compatible STT endpoint, or dictating. For push-to-talk into any text field, use the Handy app.

中文简介正在准备中。当前显示原文说明。

Use when building autonomous LLM agents operating in live environments. Protects against prompt injection, tool abuse, data exfiltration, and runaway loops. Keywords: guardrails, prompt injection, IPI, safety, sandboxing, NeMo, LLM Guard.

中文简介正在准备中。当前显示原文说明。

Audit whether all score-deciding logic in sgl-eval is vendored from NeMo-Skills, or whether some has crept into SE code. Use when the user asks "are we vendoring enough", "review vendor coverage", "audit vendoring", "is our vendoring complete", or before a release.

中文简介正在准备中。当前显示原文说明。

Add a new benchmark to sgl-eval by vendoring its NeMo-Skills dataset module and registering it in `_TABLE`. Use when the user asks to "add <benchmark>" / "support <benchmark>" / "register <benchmark>" inside the sgl-eval repo.

中文简介正在准备中。当前显示原文说明。

Kanji specialist covering character origins, radical systems, mnemonic techniques, and historical development

中文简介正在准备中。当前显示原文说明。

Vue 3 Composition API, script setup macros, reactivity system, and built-in components. Use when writing Vue SFCs, defineProps/defineEmits/defineModel, watchers, or using Transition/Teleport/Suspense/KeepAlive.

中文简介正在准备中。当前显示原文说明。