원클릭으로 Manus에서 모든 스킬 실행

sonic

스타15

포크8

업데이트2026년 7월 24일 02:55

Use when working on SONIC whole-body-control training, export, evaluation, serving, GPU routing, validation, or CUDA alignment.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

nebius

nebius/nebius-physical-ai

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

SONIC

When To Use

Use this skill for NVIDIA GEAR-SONIC whole-body-control workbench changes, including standalone training, export, evaluation, serving, image routing, and workflow composition with retargeting or MJLab.

Procedure

Confirm the current CLI surface before editing:
```
npa workbench sonic --help
```
Use train for policy training, export for export artifacts, eval for evaluation, and serve for runtime serving. Use deploy, status, and list for operational lifecycle.
Route first-party SONIC images through npa/src/npa/deploy/sonic_image_manifest.json; do not hardcode image tags in workflows or docs.
Keep S3 paths run-scoped so retraining and re-evaluation do not overwrite previous artifacts.

Three-Tier Contract

CLI: deploy, train, export, eval, serve, status, and list.
SDK/API: keep train/eval/export/serve request construction shared with service payloads and tests where possible.
YAML: SONIC workflows live under npa/src/npa/workflows/skypilot/, including sonic-train-standalone.yaml, sonic-export.yaml, sonic-eval.yaml, sonic-export-eval.yaml, and sonic-locomotion-finetuning.yaml.

Routing And Validation

Use the baked npa-sonic:0.1.2 image for L40S VM targets.
Use the host-mounted npa-sonic:0.1.2-k8s image for RTX PRO 6000 Blackwell Kubernetes targets with NVIDIA GPU Operator mounted drivers.
SONIC render validation requires RT-capable GPUs. H100 can be useful for non-render training throughput, but it is not the default render-validation target.

Gotchas

Keep SONIC_GPU_TYPE and SONIC_IMAGE_VARIANT aligned with the image manifest. Do not assume one image works across VM and Kubernetes targets.
Known issue: job ID reuse anomaly. Treat it as a deferred investigation unless the task directly targets scheduler identity handling.
CUDA 13 alignment is vendor-paced on NVIDIA x86_64 CUDA 13 and is not a Nebius-blocked item.

Verify

npa/.venv/bin/python -m pytest npa/tests/guardrails/test_skills_index.py -q

The smoke test invokes npa workbench sonic export --help and npa workbench sonic serve --help so the skill cannot regress to the older train/eval-only command list.

이 저장소의 다른 Skills

같은 저장소

dataset

nebius/nebius-physical-ai

Use when ingesting, validating, curating, or querying production sensor data as a versioned dataset-of-record, or wiring the dataset-ingest-curate workflow.

2026-07-2415

scenario-gen

nebius/nebius-physical-ai

Use when generating adversarial scenarios via RL, ranking mined failures of a policy-under-test, or wiring the adversarial-scenario-hardening workflow.

2026-07-2415

cosmos3-codebase-nav

nebius/nebius-physical-ai

Use when navigating the Cosmos3 integration in NPA or locating upstream Cosmos3 framework files, defaults, scripts, configs, recipes, and docs.

2026-07-2415

cosmos3-env-troubleshoot

nebius/nebius-physical-ai

Use when Cosmos3 setup, fetch, inference, CUDA, uv, Docker, Hugging Face, GitHub, NGC, or checkpoint staging fails in NPA or in an upstream Cosmos framework checkout.

2026-07-2415

cosmos3-setup

nebius/nebius-physical-ai

Use when setting up Cosmos3 access through NPA, checking source or Hugging Face reachability, staging the public Cosmos3 framework and checkpoint cache, or deciding which NPA workflow to use before inference.

2026-07-2415

physical-ai-context

nebius/nebius-physical-ai

Use for Claude Code reviews that need robotics, simulation, GPU-routing, sim-to-real, or BDD100K pipeline domain context.

2026-07-2415

SONIC

When To Use

Procedure

Confirm the current CLI surface before editing:

npa workbench sonic --help

Use train for policy training, export for export artifacts, eval for evaluation, and serve for runtime serving. Use deploy, status, and list for operational lifecycle.

Route first-party SONIC images through npa/src/npa/deploy/sonic_image_manifest.json; do not hardcode image tags in workflows or docs.

Keep S3 paths run-scoped so retraining and re-evaluation do not overwrite previous artifacts.

Three-Tier Contract

CLI: deploy, train, export, eval, serve, status, and list.

SDK/API: keep train/eval/export/serve request construction shared with service payloads and tests where possible.

YAML: SONIC workflows live under npa/src/npa/workflows/skypilot/, including sonic-train-standalone.yaml, sonic-export.yaml, sonic-eval.yaml, sonic-export-eval.yaml, and sonic-locomotion-finetuning.yaml.

Routing And Validation

Use the baked npa-sonic:0.1.2 image for L40S VM targets.

Use the host-mounted npa-sonic:0.1.2-k8s image for RTX PRO 6000 Blackwell Kubernetes targets with NVIDIA GPU Operator mounted drivers.

SONIC render validation requires RT-capable GPUs. H100 can be useful for non-render training throughput, but it is not the default render-validation target.

Gotchas

Keep SONIC_GPU_TYPE and SONIC_IMAGE_VARIANT aligned with the image manifest. Do not assume one image works across VM and Kubernetes targets.

Known issue: job ID reuse anomaly. Treat it as a deferred investigation unless the task directly targets scheduler identity handling.

CUDA 13 alignment is vendor-paced on NVIDIA x86_64 CUDA 13 and is not a Nebius-blocked item.

Verify

npa/.venv/bin/python -m pytest npa/tests/guardrails/test_skills_index.py -q

The smoke test invokes npa workbench sonic export --help and npa workbench sonic serve --help so the skill cannot regress to the older train/eval-only command list.

name	sonic
description	Use when working on SONIC whole-body-control training, export, evaluation, serving, GPU routing, validation, or CUDA alignment.