on CPU.
    *   Add nominal support for unsigned 16-bit integer tensor types. Note that
        very few TFLite kernels support this type natively, so its use in mobile
        ML authoring is generally discouraged.
    *   Add support for unsigned 16-bit integer tensor types in cast op.
    *   Experimental support for lowering `list_ops.tensor_list_set_item` with
        `DynamicUpdateSlice`.

   * the backport does. For now, we're keeping the class declaration in *both* flavors so that both
   * flavors can read old data or data from the other flavor. However, we strongly discourage
   * relying on this, as we have made incompatible changes to serialized forms in the past and
   * expect to do so again, as discussed in https://github.com/google/guava#important-warnings.
   */

Ha(e,a,t,s){return Sa(e,a,t,s,!1)}M.createFromInputFallback=t("value provided is not in a recognized RFC2822 or ISO format. moment construction falls back to js Date(), which is not reliable across all browsers and versions. Non RFC2822/ISO date formats are discouraged. Please refer to http://momentjs.com/guides/#/warnings/js-date/ for more info.",function(e){e._d=new Date(e._i+(e._useUTC?" UTC":""))}),M.ISO_8601=function(){},M.RFC_2822=function(){};var ja=t("moment().min is deprecated, use moment.max instead....

If a PR doesn't explain what it does or why, ask for more information.

A PR should have a specific use case that it is solving.

* If the PR is for a feature, it should have docs.
    * Unless it's a feature we want to discourage, like support for a corner case that we don't want users to use.
* The docs should include a source example file, not write Python directly in Markdown.

    airtime before a decision is made regarding whether they are to be migrated
    to the core.
*   As every PR requires several CPU/GPU hours of CI testing, we discourage
    submitting PRs to fix one typo, one warning,etc. We recommend fixing the
    same issue at the file level at least (e.g.: fix all typos in a file, fix
    all compiler warnings in a file, etc.)

 * ### IDNA Mapping and Punycode encoding
 *
 * Hostnames have different requirements and use a different encoding scheme. It consists of IDNA
 * mapping and Punycode encoding.
 *
 * In order to avoid confusion and discourage phishing attacks, [IDNA Mapping][idna] transforms
 * names to avoid confusing characters. This includes basic case folding: transforming shouting