memory

Functions

garbage_collection_cuda	Garbage collection Torch (CUDA) memory.
get_gpu_memory_map	Deprecated since version v1.5.
get_memory_profile	Deprecated since version v1.5.
get_model_size_mb	Calculates the size of a Module in megabytes.
is_cuda_out_of_memory	rtype bool
is_cudnn_snafu	rtype bool
is_oom_error	rtype bool
is_out_of_cpu_memory	rtype bool
recursive_detach	Detach all tensors in in_dict.

Utilities related to memory.

pytorch_lightning.utilities.memory.garbage_collection_cuda()

Garbage collection Torch (CUDA) memory.

Return type

None

pytorch_lightning.utilities.memory.get_gpu_memory_map()

Deprecated since version v1.5: This function was deprecated in v1.5 in favor of pytorch_lightning.accelerators.gpu._get_nvidia_gpu_stats and will be removed in v1.7.

Get the current gpu usage.

Return type

Dict[str, float]

Returns

A dictionary in which the keys are device ids as integers and values are memory usage as integers in MB.

Raises

FileNotFoundError – If nvidia-smi installation not found

pytorch_lightning.utilities.memory.get_memory_profile(mode)

Deprecated since version v1.5: This function was deprecated in v1.5 in favor of pytorch_lightning.accelerators.gpu._get_nvidia_gpu_stats and will be removed in v1.7.

Get a profile of the current memory usage.

Parameters

mode (str) –

There are two modes:

• ’all’ means return memory for all gpus

• ’min_max’ means return memory for max and min

Return type

Dict[str, float]

Returns

A dictionary in which the keys are device ids as integers and values are memory usage as integers in MB. If mode is ‘min_max’, the dictionary will also contain two additional keys:

• ’min_gpu_mem’: the minimum memory usage in MB

• ’max_gpu_mem’: the maximum memory usage in MB

pytorch_lightning.utilities.memory.get_model_size_mb(model)

Calculates the size of a Module in megabytes.

The computation includes everything in the state_dict(), i.e., by default the parameters and buffers.

Return type

float

Returns

Number of megabytes in the parameters of the input module.

pytorch_lightning.utilities.memory.recursive_detach(in_dict, to_cpu=False)

Detach all tensors in in_dict.

May operate recursively if some of the values in in_dict are dictionaries which contain instances of torch.Tensor. Other types in in_dict are not affected by this utility function.

Parameters

• in_dict (Any) – Dictionary with tensors to detach

• to_cpu (bool) – Whether to move tensor to cpu

Returns

Dictionary with detached tensors

Return type

out_dict