# For wheel verification, and uploading
auditwheel ~= 6.1.0
twine ~= 6.1.0
id
urllib3
requests

# For XLA
pyyaml

# For JAX
build ~= 1.2.2
# uv is faster than pip for installing Python packages.
uv ~= 0.5.30

# For running wheel verification script

## Base64 vs Files { #base64-vs-files }

Consider first if you can use [Request Files](../tutorial/request-files.md) for uploading binary data and [Custom Response - FileResponse](./custom-response.md#fileresponse--fileresponse-) for sending binary data, instead of encoding it in JSON.

JSON can only contain UTF-8 encoded strings, so it can't contain raw bytes.

  /**
   * Returns a cache loader that uses {@code function} to load keys, without supporting either
   * reloading or bulk loading. This allows creating a cache loader using a lambda expression.
   *
   * <p>The returned object is serializable if {@code function} is serializable.
   *
   * @param function the function to be used for loading values; must never return {@code null}

   * reloading or bulk loading. This is most useful when you can pass a lambda expression. Otherwise
   * it is useful mostly when you already have an existing function instance.
   *
   * <p>The returned object is serializable if {@code function} is serializable.
   *
   * @param function the function to be used for loading values; must never return {@code null}

 * loader and prevent it from being garbage collected. This poses a problem for environments where
 * you want to throw away the class loader. For example, dynamically reloading a web application or
 * unloading an OSGi bundle.
 *
 * <p>{@code com.google.common.base.FinalizableReferenceQueue} loads this class in its own class
 * loader. That way, this class doesn't prevent the main class loader from getting garbage

 * loader and prevent it from being garbage collected. This poses a problem for environments where
 * you want to throw away the class loader. For example, dynamically reloading a web application or
 * unloading an OSGi bundle.
 *
 * <p>{@code com.google.common.base.FinalizableReferenceQueue} loads this class in its own class
 * loader. That way, this class doesn't prevent the main class loader from getting garbage

    @Nullable V get();

    /**
     * Waits for a value that may still be loading. Unlike get(), this method can block (in the case
     * of FutureValueReference).
     *
     * @throws ExecutionException if the loading thread throws an exception
     * @throws ExecutionError if the loading thread throws an error
     */
    V waitForValue() throws ExecutionException;

    @Nullable V get();

    /**
     * Waits for a value that may still be loading. Unlike get(), this method can block (in the case
     * of FutureValueReference).
     *
     * @throws ExecutionException if the loading thread throws an exception
     * @throws ExecutionError if the loading thread throws an error
     */
    V waitForValue() throws ExecutionException;

    @Override
    public @Nullable Class<?> loadFinalizer() {
      if (disabled) {
        return null;
      }
      if (isAndroid()) {
        // We need not worry about class unloading under Android. Plus, this approach wouldn't work.
        return null;
      }
      ClassLoader systemLoader;
      try {
        systemLoader = ClassLoader.getSystemClassLoader();

    @Override
    public @Nullable Class<?> loadFinalizer() {
      if (disabled) {
        return null;
      }
      if (isAndroid()) {
        // We need not worry about class unloading under Android. Plus, this approach wouldn't work.
        return null;
      }
      ClassLoader systemLoader;
      try {
        systemLoader = ClassLoader.getSystemClassLoader();