Doc2Vec loss always showing 0

[griff4692]

class MonitorCallback(CallbackAny2Vec):
    def __init__(self, test_cui, test_sec):
        self.test_cui = test_cui
        self.test_sec = test_sec

    def on_epoch_end(self, model):
        print('Model loss:', model.get_latest_training_loss())
        for word in self.test_cui:  # show wv logic changes
             print(word, model.wv.most_similar(word))
        for word in self.test_sec:  # show dv logic changes
             print(word, model.dv.most_similar(word))

    model = Doc2Vec(vector_size=300, min_count=1, epochs=1, window=5, workers=32)
    print('Building vocab...')
    model.build_vocab(train_corpus)
    print(model.corpus_count, model.epochs)
    model.train(
        train_corpus, total_examples=model.corpus_count, compute_loss=True, epochs=model.epochs, callbacks=[monitor])
    print('Done training...')
    model.save('sec2vec.model')

Each time the callback prints, it prints 0. The second issue is that after the first epoch, the model seems pretty good according to calls to most_similar. Yet, after the second it appears random. I have a fairly large dataset so I don't think dramatic overfitting is happening. Is there a bug after the first epoch or is the learning rate getting messed up? It's tough to know what's going on because there's no within-epoch logging and the training loss is always evaluating to 0.

[griff4692]

Model loss: 0.0
Model loss: 0.0
Model loss: 0.0
Model loss: 0.0
Model loss: 0.0
Model loss: 0.0
Model loss: 0.0
Model loss: 0.0
Model loss: 0.0
Model loss: 0.0

To be clear, here is a log of the model loss

[gojomo]

As per #2983, Doc2Vec loss-tracking has never been implemented. Closing this as duplicate; will discuss further there.