Xelera Decision Tree Inference provides FPGA-accelerated inference (prediction) for real-time Classification and Regression applications when high-throughput or low-latency matters. It supports Random Forest, XGBoost and LightGBM algorithms.
- Train your own model using one of the supported frameworks (scikit-learn, XGBoost, LightGBM, H20.ai) and convert it to a unified representation (XlModel) for Alveo Accelerator Cards
- Integrate with your application via Python and run with an auto-scalable inference server
Additional resource:
- Random Forest Inference Benchmark on Lenovo Thinksystem SE350
- Blog post on accelerating Decision tree-based predictive analytics
- DRM-based licensing system
- No feature scaling required: float32-based tree traversal algorithm in FPGA
- Kernel optimized for large ensambles and RF classification (greater than hundreds of trees)
| Cards/Platform | Shell |
|---|---|
| Xilinx Alveo U50 | xilinx_u50_gen3x16_xdma_201920_3 |
| Xilinx Alveo U200 | xilinx-u200-xdma-201830.2 |
| AWS f1.2xlarge | xilinx_aws-vu9p-f1_shell-v04261818_201920_1 |
For supported features and current limitations, see supported parameters.
Xelera Decision Tree Inference is available:
See API migration for instructions to migrate from 0.3.0b3 to 0.4.0b4 release.
In case of questions, contact [email protected]