* Only kill local Gradle processes (classpath in checkout directory).
         * Not clean up global Gradle processes (i.e. classpath in ~/.gradle/...).
         * Because the step is not guaranteed to run (e.g. build timeout), we need `KILL_LEAKED_PROCESSES_FROM_PREVIOUS_BUILDS` mode.
         */
        KILL_PROCESSES_STARTED_BY_GRADLE,
        /**

          <!-- This is a required attribute and is intentionally left empty -->
          <models />
        </configuration>
        <executions>
          <execution>
            <!-- This step is required to generate xdoc, and does not generate java code -->
            <id>modello-site-doc</id>
            <goals>
              <goal>xdoc</goal>
            </goals>
            <phase>pre-site</phase>

        `(batch_size, 1)`. This enables `Model` subclasses to process scalar
        data in their `train_step()`/`test_step()`/`predict_step()` methods. \
        Note that this change may break certain subclassed models. You can
        revert back to the previous behavior by adding upranking yourself in the
        `train_step()`/`test_step()`/`predict_step()` methods, e.g. `if
        x.shape.rank == 1: x = tf.expand_dims(x, axis=-1)`. Functional models as

<code>string</code> takes the place of <code>P</code>.
And since <code>string</code> is identical to <code>string</code>,
this unification step succeeds as well.
Unification of the LHS and RHS of the equation is now finished.
Type inference succeeds because there is only one type equation,
no unification step failed, and the map is fully populated.
</p>

<p>
Unification uses a combination of <i>exact</i> and <i>loose</i>

* 🌐 Remove Vietnamese note about missing translation. PR [#9957](https://github.com/tiangolo/fastapi/pull/9957) by [@tiangolo](https://github.com/tiangolo).

### Internal

* 👷 Add GitHub Actions step dump context to debug external failures. PR [#10008](https://github.com/tiangolo/fastapi/pull/10008) by [@tiangolo](https://github.com/tiangolo).