nvForest - Highly Optimized Decision Tree Inference

nvForest is a highly-optimized and lightweight RAPIDS library that enables fast inference for decision tree models on NVIDIA GPUs and CPUs. It does not train models; it runs inference on models trained elsewhere (e.g., XGBoost, LightGBM, scikit-learn, or cuML).

nvForest uses Treelite as the common format for importing tree models. You can load a model from a file or from an in-memory scikit-learn or Treelite object, then run predictions with a scikit-learn-like API. Setting device="auto" lets you deploy the same script on machines with or without GPUs.

As an example, the following Python snippet loads an XGBoost model and runs inference on GPU:

import nvforest

# Load XGBoost model for GPU inference
fm = nvforest.load_model("/path/to/xgboost_model.ubj", device="gpu",
                         model_type="xgboost_ubj")

# Run inference (X can be a NumPy array or CuPy array)
pred = fm.predict(X)

Load a scikit-learn random forest model and get class probabilities:

import nvforest
from sklearn.ensemble import RandomForestClassifier

# Train with scikit-learn (or load a saved model)
skl_model = RandomForestClassifier(...)
skl_model.fit(X_train, y_train)

# Load into nvForest for fast GPU inference
fm = nvforest.load_from_sklearn(skl_model, device="gpu")
class_probs = fm.predict_proba(X)

For more examples and the full API, see the Getting started guide and the Python API documentation.

Supported Models

Source	Formats
XGBoost	UBJSON, JSON, legacy binary
LightGBM	Text (.txt)
scikit-learn	In-memory (RandomForest, ExtraTrees, GradientBoosting)
cuML	Via Treelite export
Treelite	Checkpoint / in-memory treelite.Model

Inference Modes

Method	Description
predict(X)	Standard predictions (class labels or regression values)
predict_proba(X)	Class probabilities (classification only)
apply(X)	Leaf indices per tree
predict_per_tree(X)	Prediction from each tree in the ensemble

You can tune performance with layout (e.g., depth_first, breadth_first) and chunk_size; use fm.optimize() to auto-tune.

---

Installation

See the RAPIDS Release Selector for the command line to install either nightly or official release nvForest packages via conda, pip, or Docker.

Build/Install from Source

See the build guide.

Contributing

We welcome contributions. For guidelines and how to get started, see the RAPIDS contributing guide.

Contact

Find out more on the RAPIDS site.

Open GPU Data Science

The RAPIDS suite of open source software libraries aims to enable execution of end-to-end data science and analytics pipelines entirely on GPUs. It relies on NVIDIA® CUDA® primitives for low-level compute optimization, exposing GPU parallelism and high-bandwidth memory through user-friendly Python interfaces.