Open afader opened 8 years ago
not right now I don't think so, good idea! What kind of use case would you have in mind with it?
@karpathy I hypothesize that the hidden state representation of text could be useful for ad hoc, information retrieval-style queries over a corpus. I want to informally test my hypothesis by getting my hands on some hidden states. Computing the hidden states from W_hh
and W_xh
seems very easy, but training the model takes time/effort. tl;dr i'm lazy :sweat_smile:
Oops, and I'm looking for the LSTM models, not vanilla RNN. So I want the other weights, too, not just the W_xh
and W_hh
.
I would also love to have the learned weights, I don't have a GPU to train with :/
I'm interested in playing around with the learned weights of a char-rnn model (referred to as
W_hh
andW_xh
in the article). Are there any pre-trained models available anywhere? I'm interested in natural language data, e.g. Wikipedia.