Closed githubrandomuser2017 closed 4 years ago
The python files in sentence_transformers/models are building block to create sentence embedding models. For an InferSent Style model, you would need an LSTM layer followed by a Pooling layer with max_pooling.
For an example, see: https://github.com/UKPLab/sentence-transformers/blob/master/examples/training/avg_word_embeddings/training_stsbenchmark_bilstm.py
These models need to be tuned on suitable data to produce meaningful sentence embeddings.
The pre-trained model page lists different pre-trained models. These models were trained on different datasets and produce meaningful sentence embeddings. They can be used without the need of any training.
Best Nils Reimers
Thank you.
I see on your documentation page that there are a variety of models we can use, e.g.
roberta-large-nli-stsb-mean-tokens
.How are these model names related to the model files found under
sentence_transformers/models/
? I see files likeBERT.py
,T5.py
, andLSTM.py
.I'm just looking to see if you have a pretrained model that uses the InferSent encoder (LSTM max pooling).