Module `Tolk_frontend.Reduce`

Reductions.

Each reduction collapses one or more axes. axis selects the axes to reduce (default: all of them) and accepts negative indices. With ~keepdim:true the reduced axes are kept as size 1 instead of being removed.

val sum : 
  ?axis:int list ->
  ?keepdim:bool ->
  ?dtype:Tolk_uop.Dtype.Val.t ->
  Tensor.t ->
  Tensor.t

sum t sums the elements of t. The accumulation is done at a widened dtype to limit overflow; pass dtype to fix it. Narrow-float inputs are cast back to their own dtype unless dtype is given.

val prod : 
  ?axis:int list ->
  ?keepdim:bool ->
  ?dtype:Tolk_uop.Dtype.Val.t ->
  Tensor.t ->
  Tensor.t

prod t multiplies the elements of t. Pass dtype to set the accumulation dtype.

val max : ?axis:int list -> ?keepdim:bool -> Tensor.t -> Tensor.t

max t is the largest element of t along the reduced axes.

val min : ?axis:int list -> ?keepdim:bool -> Tensor.t -> Tensor.t

min t is the smallest element of t along the reduced axes.

val any : ?axis:int list -> ?keepdim:bool -> Tensor.t -> Tensor.t

any t is true where at least one element along the reduced axes is non-zero.

val all : ?axis:int list -> ?keepdim:bool -> Tensor.t -> Tensor.t

all t is true where every element along the reduced axes is non-zero.