Gradient values are all zeros

[user50]

Hi
I'm encountering an issue when trying to run training on multiple GPUs using DJL. I'm using the following code:
DefaultTrainingConfig config = new DefaultTrainingConfig(loss);
config.optExecutorService(Executors.newFixedThreadPool(16));
...
EasyTrain.trainBatch(trainer, batch);
trainer.step();

However, I'm getting the following error: "Gradient values are all zeros, please call gradientCollector.backward() on your target NDArray (usually loss), before calling step(). "

using dependencies:

compile "ai.djl:api:0.19.0"
compile "ai.djl.mxnet:mxnet-engine:0.19.0"
runtimeOnly "ai.djl.mxnet:mxnet-native-auto:1.8.0"