[tutorial][benchmark] nnvm -> relay (#4368)

* [tutorial] nnvm -> relay

* use relay workload

* delete movbilenetv2 option

apps/benchmark/README.md

@@ -35,9 +35,9 @@ In general, the performance should also be good.
 
 It is recommended that you run tuning by yourself if you have your customized network or devices.
 Please follow the tutorial for
-[NVIDIA GPU](https://docs.tvm.ai/tutorials/autotvm/tune_nnvm_cuda.html),
-[ARM CPU](https://docs.tvm.ai/tutorials/autotvm/tune_nnvm_arm.html),
-[Mobile GPU](https://docs.tvm.ai/tutorials/autotvm/tune_nnvm_mobile_gpu.html).
+[NVIDIA GPU](https://docs.tvm.ai/tutorials/autotvm/tune_conv2d_cuda.html),
+[ARM CPU](https://docs.tvm.ai/tutorials/autotvm/tune_relay_arm.html),
+[Mobile GPU](https://docs.tvm.ai/tutorials/autotvm/tune_relay_mobile_gpu.html).
 
 ### NVIDIA GPU
 
@@ -67,7 +67,7 @@ python3 -m tvm.exec.rpc_tracker
 
 2. Register devices to the tracker
 * For Linux device
-  * Build tvm runtime on your device [Help](https://docs.tvm.ai/tutorials/nnvm/deploy_model_on_rasp.html#build-tvm-runtime-on-device)
+  * Build tvm runtime on your device [Help](https://docs.tvm.ai/tutorials/frontend/deploy_model_on_rasp.html#build-tvm-runtime-on-device)
   * Register your device to tracker by
   ```bash
   python3 -m tvm.exec.rpc_server --tracker=[HOST_IP]:9190 --key=[DEVICE_KEY]

apps/benchmark/arm_cpu_imagenet_bench.py

@@ -24,8 +24,7 @@ import numpy as np
 import tvm
 from tvm.contrib.util import tempdir
 import tvm.contrib.graph_runtime as runtime
-import nnvm.compiler
-import nnvm.testing
+from tvm import relay
 
 from util import get_network, print_progress
 
@@ -39,10 +38,9 @@ def evaluate_network(network, target, target_host, repeat):
     net, params, input_shape, output_shape = get_network(network, batch_size=1)
 
     print_progress("%-20s building..." % network)
-    with nnvm.compiler.build_config(opt_level=3):
-        graph, lib, params = nnvm.compiler.build(
-            net, target=target, target_host=target_host,
-            shape={'data': input_shape}, params=params, dtype=dtype)
+    with relay.build_config(opt_level=3):
+        graph, lib, params = relay.build(
+            net, target=target, target_host=target_host, params=params)
 
     tmp = tempdir()
     if 'android' in str(target):
@@ -76,7 +74,7 @@ if __name__ == "__main__":
     parser.add_argument("--network", type=str, choices=
                         ['resnet-18', 'resnet-34', 'resnet-50',
                          'vgg-16', 'vgg-19', 'densenet-121', 'inception_v3',
-                         'mobilenet', 'mobilenet_v2', 'squeezenet_v1.0', 'squeezenet_v1.1'],
+                         'mobilenet', 'squeezenet_v1.0', 'squeezenet_v1.1'],
                         help='The name of neural network')
     parser.add_argument("--model", type=str, choices=
                         ['rk3399', 'mate10', 'mate10pro', 'p20', 'p20pro',

apps/benchmark/gpu_imagenet_bench.py

@@ -23,10 +23,8 @@ import threading
 import numpy as np
 
 import tvm
-from tvm.contrib.util import tempdir
 import tvm.contrib.graph_runtime as runtime
-import nnvm.compiler
-import nnvm.testing
+from tvm import relay
 
 from util import get_network
 
@@ -34,9 +32,8 @@ from util import get_network
 def benchmark(network, target):
     net, params, input_shape, output_shape = get_network(network, batch_size=1)
 
-    with nnvm.compiler.build_config(opt_level=3):
-        graph, lib, params = nnvm.compiler.build(
-            net, target=target, shape={'data': input_shape}, params=params, dtype=dtype)
+    with relay.build_config(opt_level=3):
+        graph, lib, params = relay.build(net, target=target, params=params)
 
     # create runtime
     ctx = tvm.context(str(target), 0)
@@ -56,7 +53,7 @@ if __name__ == "__main__":
     parser.add_argument("--network", type=str, choices=
                         ['resnet-18', 'resnet-34', 'resnet-50',
                          'vgg-16', 'vgg-19', 'densenet-121', 'inception_v3',
-                         'mobilenet', 'mobilenet_v2', 'squeezenet_v1.0', 'squeezenet_v1.1'],
+                         'mobilenet', 'squeezenet_v1.0', 'squeezenet_v1.1'],
                         help='The name of neural network')
     parser.add_argument("--model", type=str,
                         choices=['1080ti', 'titanx', 'tx2', 'gfx900'], default='1080ti',

apps/benchmark/mobile_gpu_imagenet_bench.py

@@ -24,8 +24,7 @@ import numpy as np
 import tvm
 from tvm.contrib.util import tempdir
 import tvm.contrib.graph_runtime as runtime
-import nnvm.compiler
-import nnvm.testing
+from tvm import relay
 
 from util import get_network, print_progress
 
@@ -38,10 +37,9 @@ def evaluate_network(network, target, target_host, dtype, repeat):
     net, params, input_shape, output_shape = get_network(network, batch_size=1, dtype=dtype)
 
     print_progress("%-20s building..." % network)
-    with nnvm.compiler.build_config(opt_level=3):
-        graph, lib, params = nnvm.compiler.build(
-            net, target=target, target_host=target_host,
-            shape={'data': input_shape}, params=params, dtype=dtype)
+    with relay.build_config(opt_level=3):
+        graph, lib, params = relay.build(
+            net, target=target, target_host=target_host, params=params)
 
     tmp = tempdir()
     if 'android' in str(target) or 'android' in str(target_host):
@@ -75,7 +73,7 @@ if __name__ == "__main__":
     parser.add_argument("--network", type=str, choices=
                         ['resnet-18', 'resnet-34', 'resnet-50',
                          'vgg-16', 'vgg-19', 'densenet-121', 'inception_v3',
-                         'mobilenet', 'mobilenet_v2', 'squeezenet_v1.0', 'squeezenet_v1.1'],
+                         'mobilenet', 'squeezenet_v1.0', 'squeezenet_v1.1'],
                         help='The name of neural network')
     parser.add_argument("--model", type=str, choices=
                         ['rk3399'], default='rk3399',

apps/benchmark/util.py

@@ -17,7 +17,8 @@
 """Utility for benchmark"""
 
 import sys
-import nnvm
+from tvm import relay
+from tvm.relay import testing
 
 def get_network(name, batch_size, dtype='float32'):
     """Get the symbol definition and random weight of a network
@@ -46,38 +47,30 @@ def get_network(name, batch_size, dtype='float32'):
     output_shape = (batch_size, 1000)
 
     if name == 'mobilenet':
-        net, params = nnvm.testing.mobilenet.get_workload(batch_size=batch_size, dtype=dtype)
-    elif name == 'mobilenet_v2':
-        net, params = nnvm.testing.mobilenet_v2.get_workload(batch_size=batch_size, dtype=dtype)
+        net, params = testing.mobilenet.get_workload(batch_size=batch_size, dtype=dtype)
     elif name == 'inception_v3':
         input_shape = (batch_size, 3, 299, 299)
-        net, params = nnvm.testing.inception_v3.get_workload(batch_size=batch_size, dtype=dtype)
+        net, params = testing.inception_v3.get_workload(batch_size=batch_size, dtype=dtype)
     elif "resnet" in name:
         n_layer = int(name.split('-')[1])
-        net, params = nnvm.testing.resnet.get_workload(num_layers=n_layer, batch_size=batch_size, dtype=dtype)
+        net, params = testing.resnet.get_workload(num_layers=n_layer, batch_size=batch_size, dtype=dtype)
     elif "vgg" in name:
         n_layer = int(name.split('-')[1])
-        net, params = nnvm.testing.vgg.get_workload(num_layers=n_layer, batch_size=batch_size, dtype=dtype)
+        net, params = testing.vgg.get_workload(num_layers=n_layer, batch_size=batch_size, dtype=dtype)
     elif "densenet" in name:
         n_layer = int(name.split('-')[1])
-        net, params = nnvm.testing.densenet.get_workload(num_layers=n_layer, batch_size=batch_size, dtype=dtype)
+        net, params = testing.densenet.get_workload(densenet_size=n_layer, batch_size=batch_size, dtype=dtype)
     elif "squeezenet" in name:
         version = name.split("_v")[1]
-        net, params = nnvm.testing.squeezenet.get_workload(batch_size=batch_size, version=version, dtype=dtype)
-    elif name == 'custom':
-        # an example for custom network
-        from nnvm.testing import utils
-        net = nnvm.sym.Variable('data')
-        net = nnvm.sym.conv2d(net, channels=4, kernel_size=(3,3), padding=(1,1))
-        net = nnvm.sym.flatten(net)
-        net = nnvm.sym.dense(net, units=1000)
-        net, params = utils.create_workload(net, batch_size, (3, 224, 224), dtype=dtype)
+        net, params = testing.squeezenet.get_workload(batch_size=batch_size, version=version, dtype=dtype)
     elif name == 'mxnet':
         # an example for mxnet model
         from mxnet.gluon.model_zoo.vision import get_model
         block = get_model('resnet18_v1', pretrained=True)
-        net, params = nnvm.frontend.from_mxnet(block)
-        net = nnvm.sym.softmax(net)
+        net, params = relay.frontend.from_mxnet(block, shape={'data': input_shape}, dtype=dtype)
+        net = net["main"]
+        net = relay.Function(net.params, relay.nn.softmax(net.body), None, net.type_params, net.attrs)
+        net = relay.Module.from_expr(net)
     else:
         raise ValueError("Unsupported network: " + name)