Treelite Runtime

3.9.1 verified Mon Apr 27 auth: no python

Treelite runtime is the lightweight inference engine for tree models compiled by Treelite. It loads compiled shared libraries (.so/.dll) and runs predictions. Current version on PyPI is 3.9.1 (note: GitHub releases have advanced to 4.x, but treelite-runtime may lag). Released monthly.

pip install treelite-runtime

Common errors

error ModuleNotFoundError: No module named 'treelite_runtime' ↓

cause The package 'treelite-runtime' is not installed.

fix

Run: pip install treelite-runtime

error FileNotFoundError: [Errno 2] No such file or directory: 'mymodel.so' ↓

cause The compiled model file path is incorrect or not generated.

fix

Ensure the .so file exists and path is correct. Generate with treelite compiler: treelite.Model.export_lib(target='mymodel.so', ...)

error OSError: /path/to/mymodel.so: cannot open shared object file ↓

cause The shared library is missing dependencies or architecture mismatch.

fix

Check platform compatibility (e.g., compile for same architecture). Use ldd mymodel.so to verify dependencies.

Warnings

gotcha The treelite-runtime package version (3.9.1) is older than the latest treelite compiler (4.x). Ensure compatibility between compiler and runtime versions. ↓

fix Use matching versions: treelite==3.9.1 and treelite-runtime==3.9.1, or upgrade both to 4.x when treelite-runtime updates.

gotcha Predictor expects a compiled model file (e.g., .so, .dll), not a Treelite model object. Trying to pass a treelite.Model object directly raises TypeError. ↓

fix Compile the model first using treelite compiler, then pass the file path to Predictor.

gotcha The Predictor class is not thread-safe. Calling predict concurrently may cause data races. ↓

fix Create separate Predictor instances per thread, or use locks.

Imports

TreeliteRuntimeError
wrong
```
from treelite import TreeliteRuntimeError
```
correct
```
from treelite_runtime import TreeliteRuntimeError
```
Common mistake: treelite-runtime is a separate package. Import from treelite_runtime, not treelite.
Predictor
```
from treelite_runtime import Predictor
```
Main class for loading compiled models.

Quickstart

Basic usage of treelite-runtime to load a compiled model and run inference.

from treelite_runtime import Predictor
# Load a compiled model (e.g., mymodel.so)
# predictor = Predictor('./mymodel.so', nthread=1)
# result = predictor.predict(batch)
print('Predictor loaded successfully')
# Note: Requires a compiled .so file generated by treelite