void ValidateModelManifest(const ModelManifest &manifest)
ModelManifest LoadModelManifest(const std::string &path)
std::string triton_config
Triton model configuration path.
std::string triton_model
Triton model artifact path.
std::string model
Exported model path.
std::string checkpoint
Source checkpoint path.
std::string resolved_training_config
Resolved training configuration path.
std::string triton_repository
Triton repository path.
std::string dtype
Triton datatype name.
std::vector< std::string > shape
Dimensions, including symbolic values.
std::string name
Tensor name.
std::string model_name
Triton model name.
std::string model_version
Triton model version.
bool enabled
Whether Triton artifacts were exported.
std::string export_timestamp_utc
UTC export timestamp.
std::string head_name
Detection head identifier.
std::vector< Tensor > outputs
Ordered output tensor contracts.
std::string precision
Numeric precision.
std::string bundle_name
Human-readable bundle name.
std::string device
Export target device.
TritonDeployment triton
Triton deployment metadata.
std::string export_format
Model serialization format.
std::string backend
Runtime backend.
Files files
Bundle-relative files.
std::vector< std::array< float, 3 > > size_priors
Per-class length, width, and height priors.
std::vector< Tensor > inputs
Ordered input tensor contracts.
std::string size_priors_source
Provenance of the size priors.
FrozenModelConfig model
Model architecture contract.
FrozenPreprocessConfig preprocessing
Preprocessing contract.
FrozenPostprocessConfig postprocessing
Postprocessing contract.
std::array< std::array< float, 2 >, 3 > pillar_map_range
XYZ pillar-map ranges.
std::array< int, 2 > pillar_map_size
Base pillar-map dimensions.
int first_up_stride
First decoder upsampling factor.
std::vector< int > stride
Backbone strides.
std::vector< int > up_stride
Decoder upsampling strides.
std::vector< std::string > class_names
Class names in output order.
int grid_y
Output grid size along Y.
int grid_x
Output grid size along X.
int num_classes
Number of semantic classes.
float min_value
Training-time minimum.
std::string type
Normalization strategy name.
float max_value
Training-time maximum.
float epsilon
Numerical stability term.
float voxel_y
Voxel width along Y.
PointFeatureNormalizationContract point_feature_normalization
Scalar feature contract.
std::array< float, 2 > x_range
Accepted X range.
std::array< float, 2 > z_range
Accepted Z range.
int num_point_features
Features expected per point.
std::array< float, 2 > y_range
Accepted Y range.
float voxel_z
Voxel width along Z.
float voxel_x
Voxel width along X.
int max_num_points
Maximum input point count.
RuntimeDefaults runtime_defaults
Overridable runtime defaults.
std::string schema_version
Manifest schema version.
FrozenContract frozen_contract
Non-overridable inference contract.
ArtifactConfig artifact
Artifact metadata.
int max_detections
Maximum final detection count.
float nms_iou_threshold
Rotated NMS IoU threshold.
std::vector< float > nms_score_thresholds
NMS class score thresholds.
float class_score_threshold
Decoder confidence threshold.
float value_threshold
Value-threshold normalization divisor.
PointFeature point_feature
Additional point-feature defaults.
Postprocessing postprocessing
Postprocessing defaults.
Preprocessing preprocessing
Preprocessing defaults.