4 ms·
Maybe you are right from theoretical point of view, however 1.latency at inference time may be an overkill, 2.look at success of fasttext model from Facebook, i
by sergeio76 7y ago
Maybe you are right from theoretical point of view, however 1.latency at inference time may be an overkill, 2.look at success of fasttext model from Facebook, it computes vectors for words and ngrams! as well. What you will have to learn with layers and filters this model memorizes. It trains faster hence you can feed huge training data, it runs fast as well.
- misterman0 7y agoPerhaps I'm mistaken because I thought blingfire was mostly a tokenizer. But now I realize it might also be a language model. How are tokens/phrases/documents represented, computationaly? Theoretically, do they live in a vector space?