Skip to navigation
Getting Started

Support Matrix

View as Markdown

Container

ItemValue
Container imagenvcr.io/nvidia/cugraph/financial-fraud-training:3.0.0
Base imageNVIDIA cuGraph + PyG
CUDA version12.6
Python version3.10

Inference Server

ItemValue
Triton Inference Servernvcr.io/nvidia/tritonserver:26.04-py3
Triton HTTP port8000
Triton gRPC port8001
Triton metrics port8002

Triton artifacts generated by v3.0.0 training are not compatible with earlier Triton server versions.

Supported GPU Architectures

ArchitectureExample GPUsSupported
AmpereA100 40GB, A100 80GB, A6000Yes
HopperH100 80GBYes

Supported Data Formats

FormatExtensionTrainingInference
CSV.csvYesYes
Apache Parquet.parquetYesYes
Apache ORC.orcYesYes
NumPy (masks only).npyYes—

Formats can be mixed within a single dataset. Mask files (train/val/test splits) are always .npy regardless of data format.

Supported Model Kinds

kindDescription
GNN_XGBoostEdge-level fraud prediction (default)
GNN_XGBoost_NPNode-level fraud prediction
XGBoostXGBoost-only baseline (no GNN); trains directly on node feature vectors

Supported GNN Encoders

Encoderencoder value
GraphSAGEsage
Graph Attention Networkgat
Graph Transformertransformer
GeneralConvgeneral

Supported LLM Providers (Explainability)

Any OpenAI-compatible endpoint is supported:

ProviderNotes
NVIDIA NIMDefault; set NVIDIA_API_KEY
OpenAISet LLM_API_KEY, LLM_BASE_URL, LLM_MODEL
Azure OpenAISet LLM_API_KEY, LLM_BASE_URL, LLM_MODEL
vLLM (self-hosted)Set LLM_API_KEY, LLM_BASE_URL, LLM_MODEL