CVM Runtime Documentation¶
Introduction¶
The CVM Runtime library is used in @CortexLabs full-node project: CortexTheseus, working for pure deterministic AI model inference.
Python API¶
Core python API briefly involves inference and graph model.
Inference package contains the model load, inference, and free pass etc. A simple inference code likes this:
import cvm
from cvm.runtime import *
ctx = cvm.cpu()
json_str, param_bytes = cvm.utils.load_model(json_path, param_path)
net = CVMAPILoadModel(json_str, param_bytes, ctx=ctx)
...
data = cvm.ndarray.array(shape, ctx=ctx)
res = CVMAPIInference(net, data)
...
CVMAPIFreeModel(net)
And the detail methods refer to the API link.
Model graph generally used to construct the network like keras does. This module is invoked by MRT to transform the MxNet model into CVM format.