keras-team/keras · error · ValueError

Cannot interpret `dtype` argument. Expected a string or an…

Error message

Cannot interpret `dtype` argument. Expected a string or an instance of DTypePolicy. Received: dtype={identifier}

What it means

Raised by the dtype-policy resolver in keras/src/dtype_policies/__init__.py get: the dtype/policy identifier must be a string (dtype name or policy string), a DTypePolicy instance, a serializable dict, or None. The received identifier has another type (e.g. a tuple, a float, or a numpy object), so no standardization path applies and the resolver gives up.

Solutions

1. Pass a string like "float32", "mixed_bfloat16", or "float8" policy names
2. Pass an existing DTypePolicy instance directly
3. Use keras.dtype_policies.serialize/deserialize dicts for config round-trips

When it happens

Trigger: Thrown at keras/src/dtype_policies/__init__.py:112 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/c789d4fe0c35f79b. Report an issue: GitHub.

Appendix: source

Thrown at keras/src/dtype_policies/__init__.py:112

    from keras.src.dtype_policies.dtype_policy import (
        _get_quantized_dtype_policy_by_str,
    )

    if identifier is None:
        return dtype_policy.dtype_policy()
    if isinstance(identifier, DTypePolicy):
        return identifier
    if isinstance(identifier, dict):
        return deserialize(identifier)
    if isinstance(identifier, str):
        if identifier.startswith(QUANTIZATION_MODES):
            return _get_quantized_dtype_policy_by_str(identifier)
        else:
            return DTypePolicy(identifier)
    try:
        return DTypePolicy(backend.standardize_dtype(identifier))
    except:
        raise ValueError(
            "Cannot interpret `dtype` argument. Expected a string "
            f"or an instance of DTypePolicy. Received: dtype={identifier}"
        )

View on GitHub (pinned to 7a34a03db6)