keras-team/keras · error · ImportError

To export to LiteRT with the PyTorch backend, you must insta

Error message

To export to LiteRT with the PyTorch backend, you must install the `litert-torch` package. Install via: pip install litert-torch

What it means

Error "To export to LiteRT with the PyTorch backend, you must install the `litert-torch` package. Install via: pip install litert-torch" thrown in keras-team/keras.

Source

Thrown at keras/src/export/litert.py:208

def export_litert_via_torch(
    model, filepath, input_signature=None, verbose=None, **kwargs
):
    """Export Keras model to LiteRT via PyTorch backend."""
    from keras.src.backend.torch.core import device_scope
    from keras.src.backend.torch.core import get_device

    device = get_device()

    # litert_torch internally imports JAX and unconditionally enables
    # jax_enable_x64, which breaks dtype-sensitive tests elsewhere.
    # We preserve the original setting and restore it after conversion.
    with _preserve_jax_x64_state():
        try:
            import litert_torch
        except ImportError:
            raise ImportError(
                "To export to LiteRT with the PyTorch backend, "
                "you must install the `litert-torch` package. "
                "Install via: pip install litert-torch"
            )

        if device != "cpu":
            for v in model.variables:
                v.value.data = v.value.data.to("cpu")

        try:
            with device_scope("cpu"):
                if input_signature is None:
                    input_signature = get_input_signature(model)

                sample_inputs = tree.map_structure(
                    lambda x: convert_spec_to_tensor(x, replace_none_number=1),
                    input_signature,
                )

View on GitHub (pinned to 7a34a03db6)

When it happens

Trigger: Thrown at keras/src/export/litert.py:208 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/7babe050cd7898c6. Report an issue: GitHub.