deeplearning4j/deeplearning4j · error · RuntimeException

Error deserializing MultiLayerConfiguration - configuration…

Error message

Error deserializing MultiLayerConfiguration - configuration may have a custom layer, vertex or preprocessor, in pre version 1.0.0-alpha JSON format. These layers can be deserialized by first registering them with NeuralNetConfiguration.registerLegacyCustomClassesForJSON(Class...)

What it means

MultiLayerConfiguration.fromJson throws this RuntimeException when an IOException during deserialization indicates a legacy (pre-1.0.0-alpha) custom layer. Unlike the beta5 hard drop, these custom classes can still be resolved if the application first registers them via NeuralNetConfiguration.registerLegacyCustomClassesForJSON(Class...). The message tells you exactly which registration call is needed.

Solutions

  1. Call NeuralNetConfiguration.registerLegacyCustomClassesForJSON(YourCustomLayer.class) before fromJson
  2. Ensure the custom class is on the classpath with the same name as in the JSON
  3. If registration is not possible, re-create the layer config manually from the JSON content

Example fix

// before
MultiLayerConfiguration conf = MultiLayerConfiguration.fromJson(json); // RuntimeException
// after
NeuralNetConfiguration.registerLegacyCustomClassesForJSON(MyCustomLayer.class);
MultiLayerConfiguration conf = MultiLayerConfiguration.fromJson(json);
Defensive patterns

Strategy: try-catch

Validate before calling

// before deserialization, ensure registration ran:
if (!legacyClassesRegistered) NeuralNetConfiguration.registerLegacyCustomClassesForJSON(MyCustomLayer.class, MyOtherLayer.class);

Try / catch

try { return MultiLayerConfiguration.fromJson(json); } catch (RuntimeException e) { if (e.getMessage().contains("registerLegacyCustomClassesForJSON")) { NeuralNetConfiguration.registerLegacyCustomClassesForJSON(MyCustomLayer.class); return MultiLayerConfiguration.fromJson(json); } throw e; }

Prevention

When it happens

Trigger: Calling fromJson on configuration JSON containing custom layers serialized before 1.0.0-alpha, without having registered those custom classes for legacy JSON handling.

Common situations: Loading very old (2017-era) model files with custom layer subclasses; forgetting the registerLegacyCustomClassesForJSON call before deserialization; class moved/renamed so registration no longer matches.

Understand the failure class

Background: "failed to unmarshal" / json.Unmarshal errors: why parsing a response into a Go struct fails and how to fix it — this error's family across 23 libraries.

Related errors


AI-assisted analysis of deeplearning4j/deeplearning4j@4c22ac5fe4 (2026-09-07). Data as JSON: /api/errors/a37c2ac0aed12eec. Report an issue: GitHub.

Appendix: source

Thrown at deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/MultiLayerConfiguration.java:244

                    //Check for legacy custom layers: "Could not resolve type id 'CustomLayer' as a subtype of [simple type, class org.deeplearning4j.nn.conf.layers.Layer]: known type ids = [Bidirectional, CenterLossOutputLayer, CnnLossLayer, ..."
                    //1.0.0-beta5: dropping support for custom layers defined in pre-1.0.0-beta format. Built-in layers from these formats still work
                    String msg = e2.getMessage();
                    if(msg != null && msg.contains("Could not resolve type id")){
                        throw new RuntimeException("Error deserializing MultiLayerConfiguration - configuration may have a custom " +
                                "layer, vertex or preprocessor, in pre version 1.0.0-beta JSON format.\nModels in legacy format with custom" +
                                " layers should be loaded in 1.0.0-beta to 1.0.0-beta4 and saved again, before loading in the current version of DL4J", e);
                    }
                    throw new RuntimeException(e2);
                } catch (IOException e2) {
                    throw new RuntimeException(e2);
                }
            }
            throw new RuntimeException(e);
        } catch (IOException e) {
            //Check if this exception came from legacy deserializer...
            String msg = e.getMessage();
            if (msg != null && msg.contains("legacy")) {
                throw new RuntimeException("Error deserializing MultiLayerConfiguration - configuration may have a custom " +
                        "layer, vertex or preprocessor, in pre version 1.0.0-alpha JSON format. These layers can be " +
                        "deserialized by first registering them with NeuralNetConfiguration.registerLegacyCustomClassesForJSON(Class...)", e);
            }
            throw new RuntimeException(e);
        }


        //To maintain backward compatibility after loss function refactoring (configs generated with v0.5.0 or earlier)
        // Previously: enumeration used for loss functions. Now: use classes
        // IN the past, could have only been an OutputLayer or RnnOutputLayer using these enums
        int layerCount = 0;
        JsonNode confs = null;
        for (NeuralNetConfiguration nnc : conf.getConfs()) {
            Layer l = nnc.getLayer();
            if (l instanceof BaseOutputLayer && ((BaseOutputLayer) l).getLossFn() == null) {
                //lossFn field null -> may be an old config format, with lossFunction field being for the enum
                //if so, try walking the JSON graph to extract out the appropriate enum value

View on GitHub (pinned to 4c22ac5fe4)