r3ndd
r3ndd
If it's smaller than the pretrained 1.5B model then does that mean it doesn't use the pretrained model?
I just ran a test in Collaboratory using an input that is 1004 tokens in length. GPT-2 Simple, using the **355M model**, took approximately 11 seconds to generate the remaining...
Same issue, I think this project is dead.
I am having the same issue, in the meantime you can just take the code from here: https://github.com/keras-team/keras-preprocessing/blob/master/keras_preprocessing/text.py#L488
Keep in mind that these vectors are a lot stronger since the PCA vector should be more accurate (I notice that instability occurs earlier). In general I was able to...
Another option is to use the authors' unsupervised method (random pairings). That should be as fast as your old method while also creating a good amount of variance for PCA.
Just [email protected] is alright. Thanks!
@andymakhk I might be wrong about this, but it is my understanding that RNNs such as LSTM networks use some form of memory in time steps, where previous inputs influence...
@vkaracic Thanks! This was actually very helpful. The only thing I'm stuck on now is how do I reset the network's memory? For instance, if I'm generating a sentence letter-by-letter...