FM / resnetdemo

/usr/local/lib64/python3.6/site-packages/caffe2/contrib/playground/resnetdemo/

/usr/local/lib64/python3.6/site-packages/caffe2/contrib/playground/resnetdemo

Name	Size	Mode	Actions
__pycache__/	-	0755	rm
caffe2_resnet50_default_forward.py	653	0644	edit dl rm
caffe2_resnet50_default_param_update.py	1331	0644	edit dl rm
explicit_resnet_forward.py	11491	0644	edit dl rm
explicit_resnet_param_update.py	2268	0644	edit dl rm
gfs_IN1k.py	1406	0644	edit dl rm
IN1k_resnet.py	1498	0644	edit dl rm
IN1k_resnet_no_test_model.py	1880	0644	edit dl rm
override_no_test_model_no_checkpoint.py	307	0644	edit dl rm
rendezvous_filestore.py	1416	0644	edit dl rm
__init__.py	0	0644	edit dl rm

Edit: /usr/local/lib64/python3.6/site-packages/caffe2/contrib/playground/resnetdemo/IN1k_resnet.py (1498B)

import numpy as np

from caffe2.python import workspace, cnn, core
from caffe2.python import timeout_guard
from caffe2.proto import caffe2_pb2

def init_model(self):
    train_model = cnn.CNNModelHelper(
        order="NCHW",
        name="resnet",
        use_cudnn=True,
        cudnn_exhaustive_search=False
    )
    self.train_model = train_model

test_model = cnn.CNNModelHelper(
        order="NCHW",
        name="resnet_test",
        use_cudnn=True,
        cudnn_exhaustive_search=False,
        init_params=False,
    )
    self.test_model = test_model

self.log.info("Model creation completed")

def fun_per_epoch_b4RunNet(self, epoch):
    pass

def fun_per_iter_b4RunNet(self, epoch, epoch_iter):

learning_rate = 0.05
    for idx in range(self.opts['distributed']['first_xpu_id'],
                     self.opts['distributed']['first_xpu_id'] +
                     self.opts['distributed']['num_xpus']):
        caffe2_pb2_device = caffe2_pb2.CUDA if \
            self.opts['distributed']['device'] == 'gpu' else \
            caffe2_pb2.CPU
        with core.DeviceScope(core.DeviceOption(caffe2_pb2_device, idx)):
            workspace.FeedBlob(
                '{}_{}/lr'.format(self.opts['distributed']['device'], idx),
                np.array(learning_rate, dtype=np.float32)
            )

def run_training_net(self):
    timeout = 2000.0
    with timeout_guard.CompleteInTimeOrDie(timeout):
        workspace.RunNet(self.train_model.net.Proto().name)

import numpy as np from caffe2.python import workspace, cnn, core from caffe2.python import timeout_guard from caffe2.proto import caffe2_pb2 def init_model(self): train_model = cnn.CNNModelHelper( order="NCHW", name="resnet", use_cudnn=True, cudnn_exhaustive_search=False ) self.train_model = train_model test_model = cnn.CNNModelHelper( order="NCHW", name="resnet_test", use_cudnn=True, cudnn_exhaustive_search=False, init_params=False, ) self.test_model = test_model self.log.info("Model creation completed") def fun_per_epoch_b4RunNet(self, epoch): pass def fun_per_iter_b4RunNet(self, epoch, epoch_iter): learning_rate = 0.05 for idx in range(self.opts['distributed']['first_xpu_id'], self.opts['distributed']['first_xpu_id'] + self.opts['distributed']['num_xpus']): caffe2_pb2_device = caffe2_pb2.CUDA if \ self.opts['distributed']['device'] == 'gpu' else \ caffe2_pb2.CPU with core.DeviceScope(core.DeviceOption(caffe2_pb2_device, idx)): workspace.FeedBlob( '{}_{}/lr'.format(self.opts['distributed']['device'], idx), np.array(learning_rate, dtype=np.float32) ) def run_training_net(self): timeout = 2000.0 with timeout_guard.CompleteInTimeOrDie(timeout): workspace.RunNet(self.train_model.net.Proto().name)