TF_DerivedResultTypeAttr output_type = TF_DerivedResultTypeAttr<0>;
}

def TF_ArgMinOp : TF_Op<"ArgMin", [Pure]> {
  let summary = [{
Returns the index with the smallest value across dimensions of a tensor.
  }];

  let description = [{
Note that in case of ties the identity of the return value is not guaranteed.

Usage:
  ```python
  import tensorflow as tf

$fs-min;\n\n      // Base font-size formatting\n      // No need to check if the unit is valid, because we did that before\n      $min-width: if($rfs-font-size-unit == rem, #{$fs-min / $rfs-rem-value}rem, #{$fs-min}px);\n\n      // If two-dimensional, use smallest of screen width and height\n      $variable-unit: if($rfs-two-dimensional, vmin, vw);\n\n      // Calculate the variable width between 0 and $rfs-breakpoint\n      $variable-width: #{$fs-diff * 100 / $rfs-breakpoint}#{$variable-unit};\n\n      //...

$fs-min;\n\n      // Base font-size formatting\n      // No need to check if the unit is valid, because we did that before\n      $min-width: if($rfs-font-size-unit == rem, #{$fs-min / $rfs-rem-value}rem, #{$fs-min}px);\n\n      // If two-dimensional, use smallest of screen width and height\n      $variable-unit: if($rfs-two-dimensional, vmin, vw);\n\n      // Calculate the variable width between 0 and $rfs-breakpoint\n      $variable-width: #{$fs-diff * 100 / $rfs-breakpoint}#{$variable-unit};\n\n      //...

of the deepest red, by about 5-1/3 Inches at least. And this is a
farther Evidence, that the Sines of Incidence and Refraction of the
several sorts of Rays, hold the same Proportion to one another in the
smallest Refractions which they do in the greatest.

My Progress in making this nice and troublesome Experiment I have set
down more at large, that they that shall try it after me may be aware of

    *   [Improve support](https://github.com/tensorflow/tensorflow/issues/33862)
        for dynamically-sized TensorArray inside `tf.function`.
*   `tf.math`:
    *   Narrow down `argmin`/`argmax` contract to always return the smallest
        index for ties.
    *   `tf.math.reduce_variance` and `tf.math.reduce_std` return correct
        computation for complex types and no longer support integer types.