doc: update docstring and tesseract input/output handling

demo/_showcase/ansys-qoi/README.md

@@ -150,3 +150,57 @@ python main.py
 
 **Outputs:**
 - `outputs/predictions_YYYYMMDD_HHMMSS.csv`
+
+
+### GPU Acceleration
+
+By default, this workflow runs on CPU, which is sufficient for the included sample dataset (5 samples). For larger datasets or production workloads, GPU acceleration is recommended for the `qoi_train` and `qoi_inference` components.
+
+#### Prerequisites
+
+- NVIDIA GPU with CUDA support
+- NVIDIA Container Toolkit installed on the host system
+
+#### Configuration Steps
+
+**1. Update Base Image**
+
+Modify the `tesseract_config.yaml` file in both `qoi_train` and `qoi_inference` components to use a CUDA-enabled base image:
+
+```yaml
+build_config:
+  base_image: "nvidia/cuda:12.8.1-runtime-ubuntu24.04"
+  # ... rest of configuration
+```
+
+**2. Configure PyTorch with CUDA Support**
+
+Update the `scripts/pyproject.toml` file in both components to install GPU-enabled PyTorch packages:
+
+```toml
+dependencies = [
+    "torch",
+    "torchvision",
+    # ... other dependencies
+]
+
+[[tool.uv.index]]
+name = "pytorch-cu128"
+url = "https://download.pytorch.org/whl/cu128"
+
+
+[tool.uv.sources]
+torch = { index = "pytorch-cu128" }
+torchvision = { index = "pytorch-cu128" }
+```
+
+**3. Enable GPU Access During Execution**
+
+When running the workflow, use the `--gpus` flag to grant GPU access to the containers:
+
+```bash
+tesseract run --gpus all qoi_train
+tesseract run --gpus all qoi_inference
+```
+
+For more details on GPU configuration options, see the [Tesseract CLI documentation](https://docs.pasteurlabs.ai/projects/tesseract-core/latest/content/api/tesseract-cli.html#cmdoption-tesseract-run-gpus).

demo/_showcase/ansys-qoi/qoi_dataset/main.py

@@ -4,24 +4,24 @@
 
 if __name__ == "__main__":
     # Ensure output folder exists
-    local_outputs = Path(__file__).parent.resolve() / "outputs"
-    local_outputs.mkdir(parents=True, exist_ok=True)
-
-    here = Path("/tesseract/")
-    CONFIG = here / "inputs/config.yaml"
-    SIM_FOLDER = here / "inputs/Ansys_Runs"
+    here = Path(__file__).parent.resolve()
     OUTPUT_DIR = here / "outputs"
-    DATASET_FOLDER = OUTPUT_DIR / "dataset"
+    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
+
+    CONFIG = "config.yaml"
+    SIM_FOLDER = "Ansys_Runs"
+    DATASET_FOLDER = "dataset"
 
     inputs = {
-        "config": str(CONFIG),
-        "sim_folder": str(SIM_FOLDER),
-        "dataset_folder": str(DATASET_FOLDER),
+        "config": CONFIG,
+        "sim_folder": SIM_FOLDER,
+        "dataset_folder": DATASET_FOLDER,
     }
 
     qoi_dataset = Tesseract.from_image(
         "qoi_dataset",
-        volumes=["./inputs:/tesseract/inputs:ro", "./outputs:/tesseract/outputs:rw"],
+        input_path="./inputs",
+        output_path="./outputs",
     )
 
     with qoi_dataset:

demo/_showcase/ansys-qoi/qoi_dataset/scripts/process/utils.py

@@ -90,23 +90,14 @@ def sample_points_with_spheres(
 ) -> tuple[np.ndarray, np.ndarray | None]:
     """Sample points on a mesh, allocating more samples inside given spheres.
 
-    Parameters
-    ----------
-    mesh : TriangleMesh
-        Full mesh to sample from.
-    n_points : int
-        Total number of points to return.
-    method : {"poisson", "uniform"}
-        Sampling method.
-    spheres : iterable of (center, radius)
-        Each center is a (3,) np.ndarray; radius is a float.
-    sphere_fraction : float in (0, 1]
-        Fraction of total samples to dedicate to all spheres combined.
-        The remainder is sampled over the full mesh.
+    Args:
+        mesh: Input triangle mesh
+        n_points:Number of points to sample
+        method: Sampling method, either 'poisson' or 'uniform'
+        spheres : Iterable of (center, radius), each center is a (3,) np.ndarray; radius is a float.
+        sphere_fraction : Fraction of total samples to dedicate to all spheres combined
 
     Returns:
-    -------
-    tuple[np.ndarray, np.ndarray | None]
         Tuple of (points, normals) where points is (N, 3) and normals is (N, 3) or None
     """
     spheres = list(spheres)

demo/_showcase/ansys-qoi/qoi_dataset/scripts/pyproject.toml

@@ -4,8 +4,9 @@ version = "0.1.0"
 description = "Add your description here"
 requires-python = ">=3.10, <3.14"
 dependencies = [
-    "open3d>=0.18.0,<0.19.0",
-    "torch>=2.9.1",
+    "open3d",
+    "torch",
+    "pyyaml",
 ]
 
 [tool.hatch.build.targets.wheel]

demo/_showcase/ansys-qoi/qoi_dataset/tesseract_api.py

@@ -9,11 +9,14 @@
 from pydantic import BaseModel, Field
 from torch.utils._pytree import tree_map
 
+from tesseract_core.runtime.config import get_config
+from tesseract_core.runtime.experimental import InputFileReference, OutputFileReference
+
 
 class InputSchema(BaseModel):
     """Input schema for QoI dataset generation."""
 
-    config: str = Field(description="Configuration file")
+    config: InputFileReference = Field(description="Configuration file")
 
     sim_folder: str = Field(
         description="Folder path containing Ansys Fluent simulations with CAD files and QoI reports",
@@ -27,7 +30,7 @@ class InputSchema(BaseModel):
 class OutputSchema(BaseModel):
     """Output schema for QoI dataset generation."""
 
-    data: list[str | Path] = Field(
+    data: list[OutputFileReference] = Field(
         description="List of npz files containing point cloud data, simulation parameters and QoI (if available)",
     )
 
@@ -36,10 +39,14 @@ def evaluate(inputs: Any) -> Any:
     """Process simulation data and generate NPZ dataset files."""
     from process.npz import NPZProcessor
 
+    config = get_config()
+    input_base = Path(config.input_path)
+    output_base = Path(config.output_path)
+
     processor = NPZProcessor(
-        root=Path(inputs["sim_folder"]),
-        out_dir=Path(inputs["dataset_folder"]),
-        config_path=Path(inputs["config"]),
+        root=input_base / inputs["sim_folder"],
+        out_dir=output_base / inputs["dataset_folder"],
+        config_path=input_base / inputs["config"],
     )
     processed_files = processor.build()
 

demo/_showcase/ansys-qoi/qoi_dataset/tesseract_requirements.txt

@@ -1,4 +1 @@
-#torch
-open3d
-pyyaml
 ./scripts

demo/_showcase/ansys-qoi/qoi_inference/main.py

@@ -4,25 +4,24 @@
 
 if __name__ == "__main__":
     # Ensure output folder exists
-    local_outputs = Path(__file__).parent.resolve() / "outputs"
-    local_outputs.mkdir(parents=True, exist_ok=True)
+    here = Path(__file__).parent.resolve()
+    OUTPUT_DIR = here / "outputs"
+    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
 
-    here = Path("/tesseract/")
-    CONFIG = here / "inputs/config.yaml"
-    DATASET_FOLDER = here / "inputs/dataset_inference"
-    TRAINED_MODEL = here / "inputs/model.pkl"
-    SCALER = here / "inputs/scaler.pkl"
+    CONFIG = "config.yaml"
+    DATASET_FOLDER = "dataset_inference"
+    TRAINED_MODEL = "model.pkl"
+    SCALER = "scaler.pkl"
 
     inputs = {
-        "config": str(CONFIG),
-        "data_folder": str(DATASET_FOLDER),
-        "trained_model": str(TRAINED_MODEL),
-        "scaler": str(SCALER),
+        "config": CONFIG,
+        "data_folder": DATASET_FOLDER,
+        "trained_model": TRAINED_MODEL,
+        "scaler": SCALER,
     }
 
     qoi_inference = Tesseract.from_image(
-        "qoi_inference",
-        volumes=["./inputs:/tesseract/inputs:ro", "./outputs:/tesseract/outputs:rw"],
+        "qoi_inference", input_path="./inputs", output_path="./outputs"
     )
 
     with qoi_inference:

demo/_showcase/ansys-qoi/qoi_inference/scripts/pyproject.toml

@@ -4,8 +4,9 @@ version = "0.1.0"
 description = "Add your description here"
 requires-python = ">=3.10, <3.14"
 dependencies = [
-    "open3d>=0.18.0,<0.19.0",
-    "torch>=2.9.1",
+    "open3d",
+    "torch",
+    "torchvision",
     "pyyaml",
     "scikit-learn",
 ]

demo/_showcase/ansys-qoi/qoi_inference/tesseract_api.py

@@ -15,20 +15,22 @@
 from torch.utils._pytree import tree_map
 
 from tesseract_core.runtime import Array, Float32
+from tesseract_core.runtime.config import get_config
+from tesseract_core.runtime.experimental import InputFileReference
 
 
 class InputSchema(BaseModel):
     """Input schema for QoI model inference."""
 
-    config: str = Field(description="Configuration file")
+    config: InputFileReference = Field(description="Configuration file")
 
     data_folder: str = Field(
         description="Folder containing npz files with point cloud data and simulation parameters"
     )
-    trained_model: str = Field(
+    trained_model: InputFileReference = Field(
         description="Pickle file containing weights of trained model"
     )
-    scaler: str = Field(
+    scaler: InputFileReference = Field(
         description="Pickle file containing the scaling method for the dataset"
     )
 
@@ -47,8 +49,12 @@ def evaluate(inputs: Any) -> Any:
     from process.models import HybridPointCloudTreeModel
     from process.scaler import ScalingPipeline
 
-    config_path = Path(inputs["config"])
-    data_folder_path = Path(inputs["data_folder"])
+    config = get_config()
+    input_base = Path(config.input_path)
+    output_base = Path(config.output_path)
+
+    config_path = input_base / inputs["config"]
+    data_folder_path = input_base / inputs["data_folder"]
     files = [str(p.resolve()) for p in data_folder_path.glob("*.npz")]
 
     data_files = [Path(f) for f in files]
@@ -58,12 +64,8 @@ def evaluate(inputs: Any) -> Any:
     with open(inputs["config"]) as f:
         config = yaml.safe_load(f)
 
-    # Create output directory
-    output_dir = Path("/tesseract/outputs")
-    output_dir.mkdir(parents=True, exist_ok=True)
-
     # Load the scaling pipeline from saved pickle file
-    scaling_pipeline = ScalingPipeline.load(Path(inputs["scaler"]))
+    scaling_pipeline = ScalingPipeline.load(input_base / inputs["scaler"])
 
     # Get all inference samples from the dataset
     inference_samples = [raw_dataset[i] for i in range(len(raw_dataset))]
@@ -87,7 +89,7 @@ def evaluate(inputs: Any) -> Any:
     # Load the trained model
     print("Loading trained model...")
     model = HybridPointCloudTreeModel()
-    model.load(inputs["trained_model"])
+    model.load(input_base / inputs["trained_model"])
 
     # Make predictions
     print("Making predictions...")
@@ -108,7 +110,7 @@ def evaluate(inputs: Any) -> Any:
     timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
 
     # Save as CSV
-    csv_path = output_dir / f"predictions_{timestamp}.csv"
+    csv_path = output_base / f"predictions_{timestamp}.csv"
     predictions_array = qoi_predictions.numpy()
 
     # Determine number of QoI outputs

demo/_showcase/ansys-qoi/qoi_inference/tesseract_requirements.txt

@@ -1,2 +1 @@
-pyyaml
 ./scripts

demo/_showcase/ansys-qoi/qoi_train/main.py

@@ -4,17 +4,16 @@
 
 if __name__ == "__main__":
     # Ensure output folder exists
-    local_outputs = Path(__file__).parent.resolve() / "outputs"
-    local_outputs.mkdir(parents=True, exist_ok=True)
+    here = Path(__file__).parent.resolve()
+    OUTPUT_DIR = here / "outputs"
+    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
 
-    here = Path("/tesseract/")
-    CONFIG = here / "inputs/config.yaml"
-    DATASET_FOLDER = here / "inputs/dataset_reduced"
-    inputs = {"config": str(CONFIG), "data_folder": str(DATASET_FOLDER)}
+    CONFIG = "config.yaml"
+    DATASET_FOLDER = "dataset_reduced"
+    inputs = {"config": CONFIG, "data_folder": DATASET_FOLDER}
 
     qoi_train = Tesseract.from_image(
-        "qoi_train",
-        volumes=["./inputs:/tesseract/inputs:ro", "./outputs:/tesseract/outputs:rw"],
+        "qoi_train", input_path="./inputs", output_path="./outputs"
     )
 
     with qoi_train:

demo/_showcase/ansys-qoi/qoi_train/scripts/pyproject.toml

@@ -4,8 +4,9 @@ version = "0.1.0"
 description = "Add your description here"
 requires-python = ">=3.10, <3.14"
 dependencies = [
-    "open3d>=0.18.0,<0.19.0",
-    "torch>=2.9.1",
+    "open3d",
+    "torch",
+    "torchvision",
     "pyyaml",
     "scikit-learn",
 ]

demo/_showcase/ansys-qoi/qoi_train/tesseract_api.py

@@ -12,13 +12,14 @@
 from pydantic import BaseModel, Field
 from torch.utils._pytree import tree_map
 
-from tesseract_core.runtime.experimental import OutputFileReference
+from tesseract_core.runtime.config import get_config
+from tesseract_core.runtime.experimental import InputFileReference, OutputFileReference
 
 
 class InputSchema(BaseModel):
     """Input schema for QoI model training."""
 
-    config: str = Field(description="Configuration file")
+    config: InputFileReference = Field(description="Configuration file")
 
     data_folder: str = Field(
         description="Folder containing npz files containing point cloud data information, "
@@ -44,8 +45,12 @@ def evaluate(inputs: Any) -> Any:
     from process.train import train_hybrid_models
 
     # Convert all inputs to Path objects (handles strings, InputFileReference, and Path)
-    config_path = Path(inputs["config"])
-    data_folder_path = Path(inputs["data_folder"])
+    config = get_config()
+    input_base = Path(config.input_path)
+    output_base = Path(config.output_path)
+
+    config_path = input_base / inputs["config"]
+    data_folder_path = input_base / inputs["data_folder"]
     files = [str(p.resolve()) for p in data_folder_path.glob("*.npz")]
 
     data_files = [Path(f) for f in files]
@@ -64,8 +69,6 @@ def evaluate(inputs: Any) -> Any:
     )
 
     # Create output directory
-    output_dir = Path("/tesseract/outputs")
-    output_dir.mkdir(parents=True, exist_ok=True)
 
     # Create scaling pipeline from config
     scaling_pipeline = ScalingPipeline(config_path)
@@ -79,7 +82,7 @@ def evaluate(inputs: Any) -> Any:
         scaled_train, scaled_val, scaled_test
     )
 
-    model_folder = output_dir / "models"
+    model_folder = output_base / "models"
 
     hybrid_model_configs = config.get("hybrid_models", None)
     hybrid_training_config = config.get("hybrid_training", {})