XGBoost4J: Exporting to, then importing a model from stream results in wrong model format exception

[PossessedC0bra]

Hi there!

I'm currently having issues with exporting and importing an XGBoost model to/from a Stream. I export the model to a ByteArrayOutputStream with the saveModel method on ml.dmlc.xgboost4j.java.Booster :

xgboost/jvm-packages/xgboost4j/src/main/java/ml/dmlc/xgboost4j/java/Booster.java

Lines 393 to 396 in 3f38d98

	public void saveModel(OutputStream out) throws XGBoostError, IOException {
	out.write(this.toByteArray());
	out.close();
	}

At a later point i try to import the previously exported model again from an InputStream using the loadModel method on ml.dmlc.xgboost4j.java.XGBoost:

xgboost/jvm-packages/xgboost4j/src/main/java/ml/dmlc/xgboost4j/java/XGBoost.java

Lines 56 to 65 in 3f38d98

	public static Booster loadModel(InputStream in) throws XGBoostError, IOException {
	int size;
	byte[] buf = new byte[1<<20];
	ByteArrayOutputStream os = new ByteArrayOutputStream();
	while ((size = in.read(buf)) != -1) {
	os.write(buf, 0, size);
	}
	in.close();
	return Booster.loadModel(buf);
	}

Altough the Stream contains the exact same data the import fails with the following exception:

Caused by: ml.dmlc.xgboost4j.java.XGBoostError: [16:24:54] D:\a\xgboost\xgboost\src\learner.cc:785: Check failed: fi->Read(&tparam_.objective): BoostLearner: wrong model format
	at ml.dmlc.xgboost4j.java.XGBoostJNI.checkCall(XGBoostJNI.java:48)
	at ml.dmlc.xgboost4j.java.Booster.loadModel(Booster.java:81)
	at ml.dmlc.xgboost4j.java.XGBoost.loadModel(XGBoost.java:64)
	...

I recently upgraded from version 1.3.1 to 1.4.1. I never encountered this problem with version 1.3.1. I noticed an entry in the changelog mentioning changes to the export/importing of models ([jvm] Add ability to load booster direct from byte array (#6655)). Might i be using the new API wrong?

Library version: XGBoost4J 1.4.1

[trivialfis]

I think this might be relevant? #7067

[PossessedC0bra]

I think this might be relevant? #7067

I can confirm the mentioned pull request solves my issue. Thanks for your quick response!