{"owner":"NVIDIA","repo":"Isaac-GR00T","hasSkills":true,"hasMcp":false,"mcpConfig":null,"found":["CLAUDE.md"],"skills":{"CLAUDE.md":"# CLAUDE.md — Isaac GR00T N1.7\n\n## Project overview\n\nIsaac GR00T N1.7 is an open vision-language-action (VLA) model for generalized humanoid robot skills.\nThe repo contains the model, training pipeline, evaluation harness, and deployment tooling.\n\n- **Language:** Python 3.12 (dGPU, Thor, DGX Spark); Python 3.10 (Orin — see deployment dir)\n- **Package manager:** [uv](https://docs.astral.sh/uv/)\n- **Build system:** setuptools (see `pyproject.toml`)\n\n## Quick-start commands\n\n```bash\n# Install (dev mode with all extras)\nuv sync --all-extras\n\n# Lint and format (uses ruff via pre-commit)\npre-commit run --all-files\n\n# Run CPU tests\npython -m pytest tests/ -m \"not gpu\" -v --timeout=300\n\n# Run GPU tests\npython -m pytest tests/ -m gpu -v --timeout=300\n\n# Build package\nuv build\n\n# Validate lockfile\nuv lock --locked\n```\n\n## Code style\n\n- Formatter: `ruff format` (double quotes, spaces, line-length 100)\n- Linter: `ruff check` with rules E, F, I (ignores E501)\n- Config lives in `pyproject.toml` under `[tool.ruff]`\n- Run `pre-commit run --all-files` before committing\n\n## Directory layout\n\n```\ngr00t/              # Main package\n  configs/          #   Training, data, and model configs\n  data/             #   Data loading, embodiment tags, dataset processing\n  eval/             #   Evaluation (run_gr00t_server.py)\n  experiment/       #   Training pipeline (launch_finetune.py, trainer.py)\n  model/            #   Model architecture (N1.7, base, modules)\n  policy/           #   Policy inference (Gr00tPolicy, server/client)\nexamples/           # Per-embodiment example configs and READMEs\nscripts/            # Deployment, conversion, and utility scripts\n  deployment/       #   Platform install scripts (dgpu, orin, thor, spark)\ntests/              # pytest suite (markers: gpu, not gpu)\ngetting_started/    # User-facing guides and notebooks\n```\n\n## Key entry points\n\n- **Fine-tune:** `bash examples/finetune.sh --base-model-path <path> --dataset-path <path> --embodiment-tag <tag> --output-dir <dir>`\n- **Inference server:** `python gr00t/eval/run_gr00t_server.py --model-path <path> --embodiment-tag <tag>`\n- **ONNX export:** `python scripts/deployment/export_onnx_n1d7.py`\n- **TensorRT build:** `python scripts/deployment/build_trt_pipeline.py`\n- **Benchmark:** `python scripts/deployment/benchmark_inference.py`\n\n## Testing\n\n- Test markers: `gpu` (requires GPU), default is CPU-safe\n- Fixtures live in `tests/fixtures/` and `demo_data/`\n- CI runs CPU and GPU tests in separate jobs with 300s timeout\n\n## Deployment platforms\n\n- **dGPU (H100, A100, RTX):** CUDA 12.8 — install via `scripts/deployment/dgpu/install_deps.sh`, container via top-level `docker/Dockerfile` (supports x86_64 and aarch64)\n- **Jetson Orin:** CUDA 12.6 — install via `scripts/deployment/orin/install_deps.sh`, container via `scripts/deployment/orin/Dockerfile`\n- **Jetson Thor:** CUDA 13.0 — install via `scripts/deployment/thor/install_deps.sh`, container via `scripts/deployment/thor/Dockerfile`\n- **DGX Spark:** CUDA 13.0 — install via `scripts/deployment/spark/install_deps.sh`, container via `scripts/deployment/spark/Dockerfile`\n\nEach Jetson/Spark platform ships an `activate_*.sh` helper (`scripts/activate_orin.sh`, `scripts/activate_spark.sh`, `scripts/activate_thor.sh`) that exports platform-specific library paths. For dGPU, the standard `source .venv/bin/activate` is sufficient.\n"},"files":{"CLAUDE.md":"# CLAUDE.md — Isaac GR00T N1.7\n\n## Project overview\n\nIsaac GR00T N1.7 is an open vision-language-action (VLA) model for generalized humanoid robot skills.\nThe repo contains the model, training pipeline, evaluation harness, and deployment tooling.\n\n- **Language:** Python 3.12 (dGPU, Thor, DGX Spark); Python 3.10 (Orin — see deployment dir)\n- **Package manager:** [uv](https://docs.astral.sh/uv/)\n- **Build system:** setuptools (see `pyproject.toml`)\n\n## Quick-start commands\n\n```bash\n# Install (dev mode with all extras)\nuv sync --all-extras\n\n# Lint and format (uses ruff via pre-commit)\npre-commit run --all-files\n\n# Run CPU tests\npython -m pytest tests/ -m \"not gpu\" -v --timeout=300\n\n# Run GPU tests\npython -m pytest tests/ -m gpu -v --timeout=300\n\n# Build package\nuv build\n\n# Validate lockfile\nuv lock --locked\n```\n\n## Code style\n\n- Formatter: `ruff format` (double quotes, spaces, line-length 100)\n- Linter: `ruff check` with rules E, F, I (ignores E501)\n- Config lives in `pyproject.toml` under `[tool.ruff]`\n- Run `pre-commit run --all-files` before committing\n\n## Directory layout\n\n```\ngr00t/              # Main package\n  configs/          #   Training, data, and model configs\n  data/             #   Data loading, embodiment tags, dataset processing\n  eval/             #   Evaluation (run_gr00t_server.py)\n  experiment/       #   Training pipeline (launch_finetune.py, trainer.py)\n  model/            #   Model architecture (N1.7, base, modules)\n  policy/           #   Policy inference (Gr00tPolicy, server/client)\nexamples/           # Per-embodiment example configs and READMEs\nscripts/            # Deployment, conversion, and utility scripts\n  deployment/       #   Platform install scripts (dgpu, orin, thor, spark)\ntests/              # pytest suite (markers: gpu, not gpu)\ngetting_started/    # User-facing guides and notebooks\n```\n\n## Key entry points\n\n- **Fine-tune:** `bash examples/finetune.sh --base-model-path <path> --dataset-path <path> --embodiment-tag <tag> --output-dir <dir>`\n- **Inference server:** `python gr00t/eval/run_gr00t_server.py --model-path <path> --embodiment-tag <tag>`\n- **ONNX export:** `python scripts/deployment/export_onnx_n1d7.py`\n- **TensorRT build:** `python scripts/deployment/build_trt_pipeline.py`\n- **Benchmark:** `python scripts/deployment/benchmark_inference.py`\n\n## Testing\n\n- Test markers: `gpu` (requires GPU), default is CPU-safe\n- Fixtures live in `tests/fixtures/` and `demo_data/`\n- CI runs CPU and GPU tests in separate jobs with 300s timeout\n\n## Deployment platforms\n\n- **dGPU (H100, A100, RTX):** CUDA 12.8 — install via `scripts/deployment/dgpu/install_deps.sh`, container via top-level `docker/Dockerfile` (supports x86_64 and aarch64)\n- **Jetson Orin:** CUDA 12.6 — install via `scripts/deployment/orin/install_deps.sh`, container via `scripts/deployment/orin/Dockerfile`\n- **Jetson Thor:** CUDA 13.0 — install via `scripts/deployment/thor/install_deps.sh`, container via `scripts/deployment/thor/Dockerfile`\n- **DGX Spark:** CUDA 13.0 — install via `scripts/deployment/spark/install_deps.sh`, container via `scripts/deployment/spark/Dockerfile`\n\nEach Jetson/Spark platform ships an `activate_*.sh` helper (`scripts/activate_orin.sh`, `scripts/activate_spark.sh`, `scripts/activate_thor.sh`) that exports platform-specific library paths. For dGPU, the standard `source .venv/bin/activate` is sufficient.\n"},"items":[{"name":"CLAUDE.md","path":"CLAUDE.md","title":"CLAUDE.md","content":"# CLAUDE.md — Isaac GR00T N1.7\n\n## Project overview\n\nIsaac GR00T N1.7 is an open vision-language-action (VLA) model for generalized humanoid robot skills.\nThe repo contains the model, training pipeline, evaluation harness, and deployment tooling.\n\n- **Language:** Python 3.12 (dGPU, Thor, DGX Spark); Python 3.10 (Orin — see deployment dir)\n- **Package manager:** [uv](https://docs.astral.sh/uv/)\n- **Build system:** setuptools (see `pyproject.toml`)\n\n## Quick-start commands\n\n```bash\n# Install (dev mode with all extras)\nuv sync --all-extras\n\n# Lint and format (uses ruff via pre-commit)\npre-commit run --all-files\n\n# Run CPU tests\npython -m pytest tests/ -m \"not gpu\" -v --timeout=300\n\n# Run GPU tests\npython -m pytest tests/ -m gpu -v --timeout=300\n\n# Build package\nuv build\n\n# Validate lockfile\nuv lock --locked\n```\n\n## Code style\n\n- Formatter: `ruff format` (double quotes, spaces, line-length 100)\n- Linter: `ruff check` with rules E, F, I (ignores E501)\n- Config lives in `pyproject.toml` under `[tool.ruff]`\n- Run `pre-commit run --all-files` before committing\n\n## Directory layout\n\n```\ngr00t/              # Main package\n  configs/          #   Training, data, and model configs\n  data/             #   Data loading, embodiment tags, dataset processing\n  eval/             #   Evaluation (run_gr00t_server.py)\n  experiment/       #   Training pipeline (launch_finetune.py, trainer.py)\n  model/            #   Model architecture (N1.7, base, modules)\n  policy/           #   Policy inference (Gr00tPolicy, server/client)\nexamples/           # Per-embodiment example configs and READMEs\nscripts/            # Deployment, conversion, and utility scripts\n  deployment/       #   Platform install scripts (dgpu, orin, thor, spark)\ntests/              # pytest suite (markers: gpu, not gpu)\ngetting_started/    # User-facing guides and notebooks\n```\n\n## Key entry points\n\n- **Fine-tune:** `bash examples/finetune.sh --base-model-path <path> --dataset-path <path> --embodiment-tag <tag> --output-dir <dir>`\n- **Inference server:** `python gr00t/eval/run_gr00t_server.py --model-path <path> --embodiment-tag <tag>`\n- **ONNX export:** `python scripts/deployment/export_onnx_n1d7.py`\n- **TensorRT build:** `python scripts/deployment/build_trt_pipeline.py`\n- **Benchmark:** `python scripts/deployment/benchmark_inference.py`\n\n## Testing\n\n- Test markers: `gpu` (requires GPU), default is CPU-safe\n- Fixtures live in `tests/fixtures/` and `demo_data/`\n- CI runs CPU and GPU tests in separate jobs with 300s timeout\n\n## Deployment platforms\n\n- **dGPU (H100, A100, RTX):** CUDA 12.8 — install via `scripts/deployment/dgpu/install_deps.sh`, container via top-level `docker/Dockerfile` (supports x86_64 and aarch64)\n- **Jetson Orin:** CUDA 12.6 — install via `scripts/deployment/orin/install_deps.sh`, container via `scripts/deployment/orin/Dockerfile`\n- **Jetson Thor:** CUDA 13.0 — install via `scripts/deployment/thor/install_deps.sh`, container via `scripts/deployment/thor/Dockerfile`\n- **DGX Spark:** CUDA 13.0 — install via `scripts/deployment/spark/install_deps.sh`, container via `scripts/deployment/spark/Dockerfile`\n\nEach Jetson/Spark platform ships an `activate_*.sh` helper (`scripts/activate_orin.sh`, `scripts/activate_spark.sh`, `scripts/activate_thor.sh`) that exports platform-specific library paths. For dGPU, the standard `source .venv/bin/activate` is sufficient.\n","category":"root","tokens":845}]}