本文整理汇总了Python中caffe2.python.muji.Allreduce方法的典型用法代码示例。如果您正苦于以下问题：Python muji.Allreduce方法的具体用法？Python muji.Allreduce怎么用？Python muji.Allreduce使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类caffe2.python.muji的用法示例。
在下文中一共展示了muji.Allreduce方法的2个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: _add_allreduce_graph
# 需要导入模块: from caffe2.python import muji [as 别名]
# 或者: from caffe2.python.muji import Allreduce [as 别名]
def _add_allreduce_graph(model):
"""Construct the graph that performs Allreduce on the gradients."""
# Need to all-reduce the per-GPU gradients if training with more than 1 GPU
all_params = model.TrainableParams()
assert len(all_params) % cfg.NUM_GPUS == 0
# The model parameters are replicated on each GPU, get the number
# distinct parameter blobs (i.e., the number of parameter blobs on
# each GPU)
params_per_gpu = int(len(all_params) / cfg.NUM_GPUS)
with c2_utils.CudaScope(0):
# Iterate over distinct parameter blobs
for i in range(params_per_gpu):
# Gradients from all GPUs for this parameter blob
gradients = [
model.param_to_grad[p] for p in all_params[i::params_per_gpu]
]
if len(gradients) > 0:
if cfg.USE_NCCL:
model.net.NCCLAllreduce(gradients, gradients)
else:
muji.Allreduce(model.net, gradients, reduced_affix='') 示例2: build_data_parallel_model
# 需要导入模块: from caffe2.python import muji [as 别名]
# 或者: from caffe2.python.muji import Allreduce [as 别名]
def build_data_parallel_model(model, single_gpu_build_func):
if model.train:
all_loss_gradients = {} # Will include loss gradients from all GPUs
# Build the model on each GPU with correct name and device scoping
for gpu_id in range(cfg.NUM_GPUS):
with core.NameScope('gpu_{}'.format(gpu_id)):
with core.DeviceScope(muji.OnGPU(gpu_id)):
all_loss_gradients.update(
single_gpu_build_func(model))
# Add backward pass on all GPUs
model.AddGradientOperators(all_loss_gradients)
if cfg.NUM_GPUS > 1:
# Need to all-reduce the per-GPU gradients if training with more
# than 1 GPU
all_params = model.TrainableParams()
assert len(all_params) % cfg.NUM_GPUS == 0, \
'This should not happen.'
# The model parameters are replicated on each GPU, get the number
# distinct parameter blobs (i.e., the number of parameter blobs on
# each GPU)
params_per_gpu = int(len(all_params) / cfg.NUM_GPUS)
with core.DeviceScope(muji.OnGPU(cfg.ROOT_GPU_ID)):
# Iterate over distinct parameter blobs
for i in range(params_per_gpu):
# Gradients from all GPUs for this parameter blob
gradients = [
model.param_to_grad[p]
for p in all_params[i::params_per_gpu]
]
if len(gradients) > 0:
if cfg.USE_NCCL:
model.net.NCCLAllreduce(gradients, gradients)
else:
muji.Allreduce(
model.net, gradients, reduced_affix='')
for gpu_id in range(cfg.NUM_GPUS):
# After all-reduce, all GPUs perform SGD updates on their identical
# params and gradients in parallel
add_parameter_update_ops(model, gpu_id)
else:
# Testing only supports running on a single GPU
with core.NameScope('gpu_{}'.format(cfg.ROOT_GPU_ID)):
with core.DeviceScope(muji.OnGPU(cfg.ROOT_GPU_ID)):
single_gpu_build_func(model)