copy¶

nvtripy.copy(input: Tensor, device: device) → Tensor[source]¶

Copies the input tensor to the specified device.

Caution

This function cannot be used in a compiled function or nvtripy.Module because it depends on evaluating its inputs, which is not allowed during compilation.

Parameters:

input (Tensor) – [dtype=T1] Input tensor.
device (device) – The target device to copy the tensor to.

Returns:

[dtype=T1] A new tensor on the specified device.

Raises:

TripyException – If the input tensor is already on the specified device, as performing copies within the same device is currently not supported.

Return type:

Tensor

DATA TYPE CONSTRAINTS:

T1: bool, float8, float16, bfloat16, float32, int4, int8, int32, int64

Example: Copying To CPU

input = tp.Tensor([1, 2, 3], device=tp.device("gpu"))
output = tp.copy(input, device=tp.device("cpu"))

Local Variables¶

>>> input
tensor([1, 2, 3], dtype=int32, loc=gpu:0, shape=(3,))

>>> output
tensor([1, 2, 3], dtype=int32, loc=cpu:0, shape=(3,))

Example: Copying To GPU

input = tp.Tensor([1, 2, 3])
output = tp.copy(input, device=tp.device("gpu"))

Local Variables¶

>>> input
tensor([1, 2, 3], dtype=int32, loc=cpu:0, shape=(3,))

>>> output
tensor([1, 2, 3], dtype=int32, loc=gpu:0, shape=(3,))