Python cfg.NUM_GPUS屬性代碼示例

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def broadcast_parameters(model):
    """Copy parameter blobs from GPU 0 over the corresponding parameter blobs
    on GPUs 1 through cfg.NUM_GPUS - 1.
    """
    if cfg.NUM_GPUS == 1:
        # no-op if only running on a single GPU
        return

    def _do_broadcast(all_blobs):
        assert len(all_blobs) % cfg.NUM_GPUS == 0, \
            ('Unexpected value for NUM_GPUS. Make sure you are not '
             'running single-GPU inference with NUM_GPUS > 1.')
        blobs_per_gpu = int(len(all_blobs) / cfg.NUM_GPUS)
        for i in range(blobs_per_gpu):
            blobs = [p for p in all_blobs[i::blobs_per_gpu]]
            data = workspace.FetchBlob(blobs[0])
            logger.debug('Broadcasting {} to'.format(str(blobs[0])))
            for i, p in enumerate(blobs[1:]):
                logger.debug(' |-> {}'.format(str(p)))
                with c2_utils.CudaScope(i + 1):
                    workspace.FeedBlob(p, data)

    _do_broadcast(model.params)
    _do_broadcast([b + '_momentum' for b in model.TrainableParams()]) 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def GetStats(self, cur_iter, lr):
        eta_seconds = self.iter_timer.average_time * (
            cfg.SOLVER.MAX_ITER - cur_iter
        )
        eta = str(datetime.timedelta(seconds=int(eta_seconds)))
        mem_stats = c2_py_utils.GetGPUMemoryUsageStats()
        mem_usage = np.max(mem_stats['max_by_gpu'][:cfg.NUM_GPUS])
        stats = dict(
            iter=cur_iter,
            lr=float(lr),
            time=self.iter_timer.average_time,
            loss=self.smoothed_total_loss.GetMedianValue(),
            eta=eta,
            mb_qsize=int(
                np.round(self.smoothed_mb_qsize.GetMedianValue())
            ),
            mem=int(np.ceil(mem_usage / 1024 / 1024))
        )
        for k, v in self.smoothed_losses_and_metrics.items():
            stats[k] = v.GetMedianValue()
        return stats 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
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='') 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def __init__(self, **kwargs):
        # Handle args specific to the DetectionModelHelper, others pass through
        # to CNNModelHelper
        self.train = kwargs.get('train', False)
        self.num_classes = kwargs.get('num_classes', -1)
        assert self.num_classes > 0, 'num_classes must be > 0'
        for k in ('train', 'num_classes'):
            if k in kwargs:
                del kwargs[k]
        kwargs['order'] = 'NCHW'
        # Defensively set cudnn_exhaustive_search to False in case the default
        # changes in CNNModelHelper. The detection code uses variable size
        # inputs that might not play nicely with cudnn_exhaustive_search.
        kwargs['cudnn_exhaustive_search'] = False
        super(DetectionModelHelper, self).__init__(**kwargs)
        self.roi_data_loader = None
        self.losses = []
        self.metrics = []
        self.do_not_update_params = []  # Param on this list are not updated
        self.net.Proto().type = cfg.MODEL.EXECUTION_TYPE
        self.net.Proto().num_workers = cfg.NUM_GPUS * 4
        self.prev_use_cudnn = self.use_cudnn
        self.gn_params = []  # Param on this list are GroupNorm parameters 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def _CorrectMomentum(self, correction):
        """The MomentumSGDUpdate op implements the update V as

            V := mu * V + lr * grad,

        where mu is the momentum factor, lr is the learning rate, and grad is
        the stochastic gradient. Since V is not defined independently of the
        learning rate (as it should ideally be), when the learning rate is
        changed we should scale the update history V in order to make it
        compatible in scale with lr * grad.
        """
        logger.info(
            'Scaling update history by {:.6f} (new lr / old lr)'.
            format(correction))
        for i in range(cfg.NUM_GPUS):
            with c2_utils.CudaScope(i):
                for param in self.TrainableParams(gpu_id=i):
                    op = core.CreateOperator(
                        'Scale', [param + '_momentum'], [param + '_momentum'],
                        scale=correction)
                    workspace.RunOperatorOnce(op) 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def create(model_type_func, train=False, gpu_id=0):
    """Generic model creation function that dispatches to specific model
    building functions.

    By default, this function will generate a data parallel model configured to
    run on cfg.NUM_GPUS devices. However, you can restrict it to build a model
    targeted to a specific GPU by specifying gpu_id. This is used by
    optimizer.build_data_parallel_model() during test time.
    """
    model = DetectionModelHelper(
        name=model_type_func,
        train=train,
        num_classes=cfg.MODEL.NUM_CLASSES,
        init_params=train
    )
    model.only_build_forward_pass = False
    model.target_gpu_id = gpu_id
    return get_func(model_type_func)(model) 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def get_net(data_loader, name):
    logger = logging.getLogger(__name__)
    blob_names = data_loader.get_output_names()
    net = core.Net(name)
    net.type = 'dag'
    for gpu_id in range(cfg.NUM_GPUS):
        with core.NameScope('gpu_{}'.format(gpu_id)):
            with core.DeviceScope(muji.OnGPU(gpu_id)):
                for blob_name in blob_names:
                    blob = core.ScopedName(blob_name)
                    workspace.CreateBlob(blob)
                net.DequeueBlobs(
                    data_loader._blobs_queue_name, blob_names)
    logger.info("Protobuf:\n" + str(net.Proto()))

    return net 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def build_data_parallel_model(model, single_gpu_build_func):
    """Build a data parallel model given a function that builds the model on a
    single GPU.
    """
    if model.only_build_forward_pass:
        single_gpu_build_func(model)
    elif model.train:
        all_loss_gradients = _build_forward_graph(model, single_gpu_build_func)
        # Add backward pass on all GPUs
        model.AddGradientOperators(all_loss_gradients)
        if cfg.NUM_GPUS > 1:
            _add_allreduce_graph(model)
        for gpu_id in range(cfg.NUM_GPUS):
            # After allreduce, all GPUs perform SGD updates on their identical
            # params and gradients in parallel
            with c2_utils.NamedCudaScope(gpu_id):
                add_single_gpu_param_update_ops(model, gpu_id)
    else:
        # Test-time network operates on single GPU
        # Test-time parallelism is implemented through multiprocessing
        with c2_utils.NamedCudaScope(model.target_gpu_id):
            single_gpu_build_func(model) 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def __init__(self, **kwargs):
        # Handle args specific to the DetectionModelHelper, others pass through
        # to CNNModelHelper
        self.train = kwargs.get('train', False)
        self.num_classes = kwargs.get('num_classes', -1)
        assert self.num_classes > 0, 'num_classes must be > 0'
        for k in ('train', 'num_classes'):
            if k in kwargs:
                del kwargs[k]
        kwargs['order'] = 'NCHW'
        # Defensively set cudnn_exhaustive_search to False in case the default
        # changes in CNNModelHelper. The detection code uses variable size
        # inputs that might not play nicely with cudnn_exhaustive_search.
        kwargs['cudnn_exhaustive_search'] = False
        super(DetectionModelHelper, self).__init__(**kwargs)
        self.roi_data_loader = None
        self.losses = []
        self.metrics = []
        self.do_not_update_params = []  # Param on this list are not updated
        self.net.Proto().type = cfg.MODEL.EXECUTION_TYPE
        self.net.Proto().num_workers = cfg.NUM_GPUS * 4
        self.prev_use_cudnn = self.use_cudnn
        self.gn_params = []  # Param on this list are GroupNorm parameters
        self.stage_params = {}  # Param on this list are updated with scalars 

# 需要導入模塊: from detectron.core.config import cfg [as 別名]
# 或者: from detectron.core.config.cfg import NUM_GPUS [as 別名]
def __init__(
        self,
        roidb,
        num_loaders=4,
        minibatch_queue_size=64,
        blobs_queue_capacity=8
    ):
        self._roidb = roidb
        self._lock = threading.Lock()
        self._perm = deque(range(len(self._roidb)))
        self._cur = 0  # _perm cursor
        # The minibatch queue holds prepared training data in host (CPU) memory
        # When training with N > 1 GPUs, each element in the minibatch queue
        # is actually a partial minibatch which contributes 1 / N of the
        # examples to the overall minibatch
        self._minibatch_queue = Queue.Queue(maxsize=minibatch_queue_size)
        self._blobs_queue_capacity = blobs_queue_capacity
        # Random queue name in case one instantiates multple RoIDataLoaders
        self._loader_id = uuid.uuid4()
        self._blobs_queue_name = 'roi_blobs_queue_{}'.format(self._loader_id)
        # Loader threads construct (partial) minibatches and put them on the
        # minibatch queue
        self._num_loaders = num_loaders
        self._num_gpus = cfg.NUM_GPUS
        self.coordinator = Coordinator()

        self._output_names = get_minibatch_blob_names()
        self._shuffle_roidb_inds()
        self.create_threads()