keras-team/keras · error · TypeError

Unable to serialize {obj} to JSON. Unrecognized type {type(o

Error message

Unable to serialize {obj} to JSON. Unrecognized type {type(obj)}.

What it means

get_json_type is the fallback serializer behind model.to_json(); after handling dicts, lists, tuples, numpy scalars, enums, and bytes, any unrecognized object type raises this TypeError naming the object and its type. It is the catch-all for 'this config value cannot be written as JSON'.

Source

Thrown at keras/src/legacy/saving/json_utils.py:218

            )
    if tf.available and isinstance(obj, tf.__internal__.CompositeTensor):
        spec = tf.type_spec_from_value(obj)
        tensors = []
        for tensor in tf.nest.flatten(obj, expand_composites=True):
            tensors.append((tensor.dtype.name, tensor.numpy().tolist()))
        return {
            "class_name": "CompositeTensor",
            "spec": get_json_type(spec),
            "tensors": tensors,
        }

    if isinstance(obj, enum.Enum):
        return obj.value

    if isinstance(obj, bytes):
        return {"class_name": "__bytes__", "value": obj.decode("utf-8")}

    raise TypeError(
        f"Unable to serialize {obj} to JSON. Unrecognized type {type(obj)}."
    )

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Locate the object named in the message and convert it to a primitive in the custom layer's get_config
2. Ensure custom get_config returns only str/int/float/bool/list/dict values
3. Use model.save('model.keras'), which supports richer Python objects

Example fix

# before
class MyLayer(layers.Layer):
    def get_config(self):
        return {'fn': self._fn}  # callable -> TypeError
# after
    def get_config(self):
        return {'fn_name': self._fn.__name__}

import json
cfg = layer.get_config()
json.dumps(cfg)  # dry-run: raises before saving if non-serializable

def config_json_safe(cfg):
    try:
        json.dumps(cfg)
        return True
    except TypeError:
        return False

try:
    model.to_json()
except TypeError as e:
    if 'Unrecognized type' not in str(e):
        raise
    model.save('model.keras')

When it happens

Trigger: model.to_json() when a layer's config holds a non-JSON value: a custom class instance, a function reference, or a type the serializer never learned.

Common situations: Custom layers whose get_config returns arbitrary Python objects; passing callables or non-primitive constants as layer arguments.

Related errors

• Layer '{self.name}' was never built and thus it doesn't have
• Data not JSON Serializable: {data}
• Targets not JSON Serializable: {targets}
• Unable to serialize {obj} to JSON, because the TypeSpec clas
• Method `compute_output_shape()` of layer {self.__class__.__n

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/653fe5dd9cd19fc8. Report an issue: GitHub.