Reduction ops
Unreleased native NuTorch shell reference. Generated by
scripts/gen-ops-reference.ts; every op also documents itself with
torch <op> --help. Tensor operands are native values, not strings.
sum
sum over all elements, or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch sum <t1> [--dim <Int>] [--keepdim]
mean
mean over all elements, or along —dim (float32, v1 fidelity)
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch mean <t1> [--dim <Int>] [--keepdim]
prod
product over all elements, or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch prod <t1> [--dim <Int>] [--keepdim]
amax
max values over all elements, or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch amax <t1> [--dim <Int>] [--keepdim]
amin
min values over all elements, or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch amin <t1> [--dim <Int>] [--keepdim]
max
max of all elements; with —dim also returns indices
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch max <t1> [--dim <Int>] [--keepdim]
min
min of all elements; with —dim also returns indices
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch min <t1> [--dim <Int>] [--keepdim]
median
median of all elements; with —dim also returns indices
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch median <t1> [--dim <Int>] [--keepdim]
argmax
index of the max, overall or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch argmax <t1> [--dim <Int>] [--keepdim]
argmin
index of the min, overall or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch argmin <t1> [--dim <Int>] [--keepdim]
all
true if all elements are true (Bool tensor)
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch all <t1> [--dim <Int>] [--keepdim]
any
true if any element is true (Bool tensor)
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch any <t1> [--dim <Int>] [--keepdim]
std
standard deviation (—correction, default 1)
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch std <t1> [--dim <Int>] [--keepdim] [--correction <Int>]
var
variance (—correction, default 1)
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch var <t1> [--dim <Int>] [--keepdim] [--correction <Int>]
nansum
sum treating NaN as zero
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch nansum <t1> [--dim <Int>] [--keepdim]
logsumexp
log(sum(exp(x))) along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch logsumexp <t1> [--dim <Int>] [--keepdim]
count_nonzero
count of nonzero elements, overall or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch count_nonzero <t1> [--dim <Int>]
cumsum
cumulative sum along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch cumsum <t1> [--dim <Int>]
cumprod
cumulative product along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch cumprod <t1> [--dim <Int>]
norm
p-norm (—p default 2), overall or along —dim
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch norm <t1> [--p <Float>] [--dim <Int>] [--keepdim]
diff
forward differences along —dim (default last)
Signature (placeholders describe arguments; this is not a runnable example):
use torch
torch diff <t1> [--dim <Int>]