一键在 Manus 中运行任何 Skill

gaponet-inf

星标18

分支3

更新时间2026年5月31日 06:43

Run GapONet inference and evaluation. Use when the user invokes /gaponet-inf or $gaponet-inf, or asks to export a GapONet checkpoint to TorchScript/JIT, run lightweight inference, deploy a policy, evaluate test motion data, or produce inference metrics.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

jiemingcui

jiemingcui/gaponet

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

GapONet Inference

Run the inference stage from the GapONet repository root, identified by scripts/run_gaponet_job.py, scripts/rsl_rl/inference_jit.py, and configs/deploy_default.json.

Workflow

Ensure data is staged. Prefer configs/deploy_default.json, which should point at the data produced by $gaponet-data.
Decide whether export is needed:
- If the user provides a JIT policy or model/policy.pt exists, run deploy directly.
- If the user provides a training checkpoint such as model/model_17950.pt, export first, then deploy.
- If neither exists, inspect runs/train/training_metrics.json or logs/ for the latest checkpoint; if none is available, ask for a checkpoint or JIT model path.
Export a training checkpoint to JIT when needed:

python scripts/run_gaponet_job.py --mode export --checkpoint <checkpoint.pt> --output-dir ./runs/inf/export

Run lightweight inference and evaluation:

python scripts/run_gaponet_job.py --mode deploy --config configs/deploy_default.json --output-dir ./runs/inf/deploy

If using the JIT model from the export step, pass it explicitly:

python scripts/run_gaponet_job.py --mode deploy --checkpoint ./runs/inf/export/policy.pt --input-data ./source/sim2real/sim2real/tasks/humanoid_operator/motions/motion_amass/edited_27dof/test.npz --output-dir ./runs/inf/deploy

Notes

Export requires Isaac Sim because it creates the task environment to infer policy dimensions.
Deploy is lightweight and does not require Isaac Sim, but it does require PyTorch, a JIT model, and test .npz data.
Standard deploy artifacts are run_manifest.json, eval_metrics.json, stdout.log, and stderr.log.

Completion

Consider inference complete only when runs/inf/deploy/run_manifest.json reports "status": "success" and eval_metrics.json exists. Report payload-grouped metrics when present.

同仓库更多 Skills

同仓库

gaponet

jiemingcui/gaponet

GapONet sim-to-real humanoid robot control workflows for Isaac Lab. Use when working on the GapONet repository or tasks involving GapONet setup, environment validation, training, evaluation, checkpoint export, lightweight deployment, DeepONet, Transformer or MLP policy configs, humanoid operator or AMASS tasks, motion data schemas, robot asset integration, or agent-friendly run artifacts.

2026-05-3118

gaponet-data

jiemingcui/gaponet

Run GapONet data preparation. Use when the user invokes /gaponet-data or $gaponet-data, or asks to process, validate, stage, copy, symlink, or prepare GapONet motion .npz data for training or inference.

2026-05-3118

gaponet-train

jiemingcui/gaponet

Run GapONet training. Use when the user invokes /gaponet-train or $gaponet-train, or asks to train a GapONet humanoid policy after data preparation using Isaac Lab, Isaac Sim, rsl_rl, DeepONet, Transformer, MLP, or the unified runner.

2026-05-3118

name	gaponet-inf
description	Run GapONet inference and evaluation. Use when the user invokes /gaponet-inf or $gaponet-inf, or asks to export a GapONet checkpoint to TorchScript/JIT, run lightweight inference, deploy a policy, evaluate test motion data, or produce inference metrics.

GapONet Inference

Run the inference stage from the GapONet repository root, identified by scripts/run_gaponet_job.py, scripts/rsl_rl/inference_jit.py, and configs/deploy_default.json.

Workflow

Ensure data is staged. Prefer configs/deploy_default.json, which should point at the data produced by $gaponet-data.
Decide whether export is needed:
- If the user provides a JIT policy or model/policy.pt exists, run deploy directly.
- If the user provides a training checkpoint such as model/model_17950.pt, export first, then deploy.
- If neither exists, inspect runs/train/training_metrics.json or logs/ for the latest checkpoint; if none is available, ask for a checkpoint or JIT model path.
Export a training checkpoint to JIT when needed:

python scripts/run_gaponet_job.py --mode export --checkpoint <checkpoint.pt> --output-dir ./runs/inf/export

Run lightweight inference and evaluation:

python scripts/run_gaponet_job.py --mode deploy --config configs/deploy_default.json --output-dir ./runs/inf/deploy

If using the JIT model from the export step, pass it explicitly:

python scripts/run_gaponet_job.py --mode deploy --checkpoint ./runs/inf/export/policy.pt --input-data ./source/sim2real/sim2real/tasks/humanoid_operator/motions/motion_amass/edited_27dof/test.npz --output-dir ./runs/inf/deploy

Notes

Export requires Isaac Sim because it creates the task environment to infer policy dimensions.
Deploy is lightweight and does not require Isaac Sim, but it does require PyTorch, a JIT model, and test .npz data.
Standard deploy artifacts are run_manifest.json, eval_metrics.json, stdout.log, and stderr.log.

Completion

Consider inference complete only when runs/inf/deploy/run_manifest.json reports "status": "success" and eval_metrics.json exists. Report payload-grouped metrics when present.