__version__ attribute missing in mode config.

https://github.com/UKPLab/sentence-transformers/blob/43e89d9d974478b85b8ef3c73006f177df64dd9c/sentence_transformers/SentenceTransformer.py#L73

I’m trying to load a model that I trained from scratch, and it appears that the SentenceTransformer.py code above is requesting a field __version__ in the model config that doesn’t appear to be there.

I haven’t looked too deeply, but in case it matters, this is a RoBERTa large NLI model (I notice in the Bert-base model I trained it’s also missing).

Also, thanks for such a great toolkit!

[nreimers]

Hi @yakazimir if you want to use your own, custom BERT model, you must load it via:

# Use BERT for mapping tokens to embeddings
word_embedding_model = models.BERT('path/to/your/bert/model')

# Apply mean pooling to get one fixed sized sentence vector
pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension(),
                               pooling_mode_mean_tokens=True,
                               pooling_mode_cls_token=False,
                               pooling_mode_max_tokens=False)

model = SentenceTransformer(modules=[word_embedding_model, pooling_model])

Best Nils Reimers

[hwaking]

邮件已收到！