Skip to content

SumReduce — fwd (cuda, torch.float32, wrapper, default)

← Back to compiled benchmarks

Result

Field Value
Operator SumReduce
Category linop
Method fwd
Device cuda
Dtype torch.float32
Compile level wrapper
Compile config default
Status ok
Class works
Graph breaks 0
Recompiles —
Eager time 19.90 us
Compiled time 20.64 us
Speedup 0.96x
Max abs diff 2.384185791015625e-07
Max rel diff 1.1138030493254364e-07
Degraded False

Tensors

Tensor Shape Dtype Device Numel Bytes View Contiguous Stride
input [5, 2, 3] torch.float32 cuda:0 30 120.00 B False True [6, 3, 1]
output [5, 2] torch.float32 cuda:0 10 40.00 B False True [2, 1]

Notes

—