Loading 3.8.3 Doc2Vec models into 4.4

[bnomis]

Loading Doc2Vec models made with version 3.8.3 into version 4.4 does not work. The error (so far) is: load exception: 'Doc2Vec' object has no attribute 'dv'

Is this something that should work?

If so I am happy to work on a patch.

Thanks

[gojomo]

Can you trigger/capture a more-detailed exception stack?

Was the model fully created in gensim-3.8.3, and can the problem be demonstrated with a minimal model created in 3.8.3?

Does the same model showing the problem load successfully in any gensim 4.0<v<4.4, or fail for all 4.x?

[bnomis]

Yes. The model was created in gensim-3.8.3.

The same issue is shown in 4.3 and 4.4 - I cannot get the earlier 4.X versions to build on my mac

The detailed stack trace is:

$ poetry run gstester --model-path=~/src/gensim/versions/release-3.8.3/gensim/gensim/test/test_data/old_d2v_models/d2v_3.8.3.mdl
2025-11-06 16:33:39 INFO     simonb gensim.utils utils.py/load/481: loading Doc2Vec object from ~/src/gensim/versions/release-3.8.3/gensim/gensim/test/test_data/old_d2v_models/d2v_3.8.3.mdl
2025-11-06 16:33:39 INFO     simonb gensim.utils utils.py/_load_specials/515: loading vocabulary recursively from ~/src/gensim/versions/release-3.8.3/gensim/gensim/test/test_data/old_d2v_models/d2v_3.8.3.mdl.vocabulary.* with mmap=None
2025-11-06 16:33:39 INFO     simonb gensim.utils utils.py/_load_specials/515: loading trainables recursively from ~/src/gensim/versions/release-3.8.3/gensim/gensim/test/test_data/old_d2v_models/d2v_3.8.3.mdl.trainables.* with mmap=None
2025-11-06 16:33:39 INFO     simonb gensim.utils utils.py/_load_specials/515: loading wv recursively from ~/src/gensim/versions/release-3.8.3/gensim/gensim/test/test_data/old_d2v_models/d2v_3.8.3.mdl.wv.* with mmap=None
2025-11-06 16:33:39 INFO     simonb gensim.utils utils.py/_load_specials/515: loading docvecs recursively from ~/src/gensim/versions/release-3.8.3/gensim/gensim/test/test_data/old_d2v_models/d2v_3.8.3.mdl.docvecs.* with mmap=None
2025-11-06 16:33:39 ERROR    simonb gensim.models.doc2vec doc2vec.py/load/811: Model load error. Was model saved using code from an older Gensim version? Try loading older model using gensim-3.8.3, then re-saving, to restore compatibility with current code.
2025-11-06 16:33:39 ERROR    simonb gensim-test gstester.py/gensim_test/18: load exception: 'Doc2Vec' object has no attribute 'dv'
Traceback (most recent call last):
  File "/Users/simonb/src/gensim/test/src/gensim_test/gstester.py", line 16, in gensim_test
    gensim.models.doc2vec.Doc2Vec.load(*args, **kwargs)
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/models/doc2vec.py", line 815, in load
    raise ae
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/models/doc2vec.py", line 809, in load
    return super(Doc2Vec, cls).load(*args, rethrow=True, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/models/word2vec.py", line 1960, in load
    raise ae
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/models/word2vec.py", line 1953, in load
    model = super(Word2Vec, cls).load(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/utils.py", line 486, in load
    obj._load_specials(fname, mmap, compress, subname)
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/models/word2vec.py", line 1969, in _load_specials
    super(Word2Vec, self)._load_specials(*args, **kwargs)
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/utils.py", line 517, in _load_specials
    getattr(self, attrib)._load_specials(cfname, mmap, compress, subname)
    ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/utils.py", line 1521, in new_func1
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/simonb/src/gensim/test/.venv/lib/python3.12/site-packages/gensim/models/doc2vec.py", line 328, in docvecs
    return self.dv
           ^^^^^^^
AttributeError: 'Doc2Vec' object has no attribute 'dv'

The code (run in the test directory of 3.8.3) I used to make the test model is:

#!/usr/bin/env python
import numpy as np
from gensim.test.utils import datapath, common_texts
from gensim import utils
from gensim.models import doc2vec


class DocsLeeCorpus(object):
    def __init__(self, string_tags=False, unicode_tags=False):
        self.string_tags = string_tags
        self.unicode_tags = unicode_tags

    def _tag(self, i):
        if self.unicode_tags:
            return "_\xa1_%d" % i
        elif self.string_tags:
            return "_*%d" % i
        return i

    def __iter__(self):
        with open(datapath("lee_background.cor")) as f:
            for i, line in enumerate(f):
                yield doc2vec.TaggedDocument(
                    utils.simple_preprocess(line), [self._tag(i)]
                )


def models_equal(model, model2):
    # check words/hidden-weights
    assert len(model.wv.vocab) == len(model2.wv.vocab)
    assert np.allclose(model.wv.vectors, model2.wv.vectors)
    if model.hs:
        assert np.allclose(model.trainables.syn1, model2.trainables.syn1)
    if model.negative:
        assert np.allclose(model.trainables.syn1neg, model2.trainables.syn1neg)
    # check docvecs
    assert len(model.docvecs.doctags) == len(model2.docvecs.doctags)
    assert (
        len(model.docvecs.offset2doctag) == len(model2.docvecs.offset2doctag)
    )
    return True

# makes a bigger model
def make_and_save_model_lee():
    corpus = DocsLeeCorpus()
    model = doc2vec.Doc2Vec(vector_size=100, min_count=2, epochs=20, workers=1)
    model.build_vocab(corpus)
    assert model.docvecs.vectors_docs.shape == (300, 100)
    model.train(corpus, total_examples=model.corpus_count, epochs=model.epochs)

    # save
    path = datapath("old_d2v_models/d2v_3.8.3.mdl")
    model.save(path)

    # load and check
    model2 = doc2vec.Doc2Vec.load(path)
    assert models_equal(model, model2)


def make_and_save_model():
    sentences = [doc2vec.TaggedDocument(words, [i]) for i, words in enumerate(common_texts)]
    vector_size = 5
    model = doc2vec.Doc2Vec(vector_size=vector_size, min_count=2, epochs=20, workers=1)
    model.build_vocab(sentences)
    assert model.docvecs.vectors_docs.shape == (len(sentences), vector_size)
    model.train(sentences, total_examples=model.corpus_count, epochs=model.epochs)

    # save
    path = datapath("old_d2v_models/d2v_3.8.3.mdl")
    model.save(path)

    # load and check
    model2 = doc2vec.Doc2Vec.load(path)
    assert models_equal(model, model2)


if __name__ == "__main__":
    make_and_save_model()

I've attached the test model.

d2v_3.8.3.mdl.zip

Thank you!

[bnomis]

@gojomo any update on this issue? Should I work on a patch? Thanks.