| 12345678910111213141516171819202122232425262728293031323334353637383940414243 |
- import numpy as np
- from caffe2.python import core, workspace
- from caffe2.quantization.server import dnnlowp_pybind11 # type: ignore[attr-defined]
- net = core.Net("test_net")
- X = np.array([[1, 2], [3, 4]]).astype(np.float32)
- W = np.array([[5, 6], [7, 8]]).astype(np.float32)
- b = np.array([0, 1]).astype(np.float32)
- workspace.FeedBlob("X", X)
- workspace.FeedBlob("W", W)
- workspace.FeedBlob("b", b)
- Y = net.FC(["X", "W", "b"], ["Y"])
- dnnlowp_pybind11.ObserveMinMaxOfOutput("test_net.minmax", 1)
- workspace.CreateNet(net)
- workspace.RunNet(net)
- print(workspace.FetchBlob("Y"))
- workspace.ResetWorkspace()
- workspace.FeedBlob("X", X)
- workspace.FeedBlob("W", W)
- workspace.FeedBlob("b", b)
- dnnlowp_pybind11.ObserveHistogramOfOutput("test_net.hist", 1)
- workspace.CreateNet(net)
- workspace.RunNet(net)
- workspace.FeedBlob("X", X)
- workspace.FeedBlob("W", W)
- workspace.FeedBlob("b", b)
- dnnlowp_pybind11.AddOutputColumnMaxHistogramObserver(
- net._net.name, "test_net._col_max_hist", ["Y"]
- )
- workspace.RunNet(net)
|
