GitHub - kossisoroyce/timber: Ollama for classical ML models. AOT compiler that turns XGBoost, LightGBM, scikit-learn, CatBoost & ONNX models into native C99 inference code. One command to load, one command to serve. 336x faster than Python inference.

$ pip install timber-compiler
$ timber serve https://raw.githubusercontent.com/kossisoroyce/timber/main/examples/breast_cancer_model.json

  ✓ Downloaded              breast_cancer_model.json
  ✓ Format detected         xgboost
  ✓ Parsed model            50 trees · 30 features · binary:logistic
  ✓ Optimized               3/5 passes applied
  ✓ Generated C99           169 lines
  ✓ Compiled binary         47.9 KB

  Serving    breast_cancer_model
  Endpoint   http://localhost:11434
  Framework  xgboost  ·  50 trees  ·  30 features

  POST  http://localhost:11434/api/predict
  GET   http://localhost:11434/api/models
  GET   http://localhost:11434/api/health

$ curl -s http://localhost:11434/api/predict \
    -H 'Content-Type: application/json' \
    -d '{"model": "breast_cancer_model", "inputs": [[1.799e+01, 1.038e+01, 1.228e+02, 1.001e+03, 0.1184, 0.2776, 0.3001, 0.1471, 0.2419, 0.07871, 1.095, 0.9053, 8.589, 153.4, 0.006399, 0.04904, 0.05373, 0.01587, 0.03003, 0.006193, 2.538e+01, 1.733e+01, 1.846e+02, 2.019e+03, 0.1622, 0.6656, 0.7119, 0.2654, 0.4601, 0.1189]]}'

{"model": "breast_cancer_model", "outputs": [[0.9971]], "n_samples": 1}

$ timber load fraud_model.json --name fraud-detector
$ timber serve fraud-detector

  ┌─────────────────────────────────────────────────────────┐
  │                     timber load                         │
  │                                                         │
  │  Model file  ──►  Parser  ──►  Timber IR  ──►  Optimizer│
  │  (.json/.pkl/                  (typed AST)   (dead-leaf  │
  │   .txt/.onnx)                               elim, quant, │
  │                                              branch-sort) │
  │                                     │                    │
  │                                     ▼                    │
  │                               C99 Emitter                │
  │                                     │                    │
  │                    ┌────────────────┼────────────────┐   │
  │                    ▼                ▼                ▼   │
  │               model.c         model.h        model_data.c│
  │               (inference)     (public API)   (tree data)  │
  │                    │                                     │
  │                    └──► gcc / clang ──► model.so         │
  └─────────────────────────────────────────────────────────┘
                              │
                              ▼
                      timber serve <name>
                   http://localhost:11434/api/predict

pip install timber-compiler

timber serve https://yourhost.com/models/fraud_model.json

timber load fraud_model.json --name fraud-detector
timber serve fraud-detector

curl -s http://localhost:11434/api/predict \
  -H "Content-Type: application/json" \
  -d '{"model": "fraud-detector", "inputs": [[1.2, 0.4, 3.1, 0.9]]}'

{"model": "fraud-detector", "outputs": [[0.031]], "latency_us": 1.8}

python benchmarks/run_benchmarks.py --output benchmarks/results.json
python benchmarks/render_table.py   --input  benchmarks/results.json

curl -s http://localhost:11434/api/predict \
  -H "Content-Type: application/json" \
  -d '{
    "model": "fraud-detector",
    "inputs": [
      [1.2, 0.4, 3.1, 0.9],
      [0.1, 2.3, 1.0, 4.4]
    ]
  }'

timber load   <path> --name <name>   Compile and register a model
timber serve  <name> [--port N]      Start the inference server
timber list                          List registered models
timber inspect <name>                Show model IR summary and schema
timber validate <name>               Run numerical validation against source
timber bench  <name>                 Benchmark latency and throughput
timber pull   <url>  --name <name>   Download and compile from URL
timber remove <name>                 Remove a model from the registry

python examples/quickstart_xgboost.py   # trains, compiles, and benchmarks
python examples/quickstart_lightgbm.py
python examples/quickstart_sklearn.py

git clone https://github.com/kossisoroyce/timber.git
cd timber
pip install -e ".[dev]"
pytest tests/ -v                    # 146 tests
ruff check timber/                  # linting

@misc{royce2026timber,
  title        = {Timber: Compiling Classical Machine Learning Models to Native Inference Binaries},
  author       = {Kossiso Royce},
  year         = {2026},
  howpublished = {GitHub repository and technical paper},
  institution  = {Electricsheep Africa},
  url          = {https://github.com/kossisoroyce/timber}
}