search

karpathy / char-rnn

Multi-layer Recurrent Neural Networks (LSTM, GRU, RNN) for character-level language models in Torch
11.58k stars 2.58k forks source link

Question about saving models when we have clones of protos #190

Open jpilaul opened 7 years ago

jpilaul commented 7 years ago

Hey, This is a general question that I was hoping someone can help me answer. I am trying to load a saved model from this repo https://github.com/cheng6076/SNLI-attention which uses a lot of the same model_utils as in this repo. When I load the model, the test accuracy is around 33%, as if the model was not trained (should be 83%).

Here is my question:

You are saving protos but I don't understand how you get this from the updates that are made on the cloned protos (https://github.com/karpathy/char-rnn/blob/master/train.lua#L202)

Thanks