textgenrnn
textgenrnn copied to clipboard
minimaxir
Easily train your own text-generating neural network of any size and complexity on any text dataset with a few lines of code.
Hello! I am going to find why after several epochs textgenrnn is unable to generate text. It happens only with new_model=True Here at 6-th epoch dramatically decrease of loss and...
Is it possible to add a prompt to the dataset so for example I could have a question in the data as the prompt and a desired response as the...
Is there a way to use a tensor processing unit for acceleration? If not, is this feature going to be added in the future?
Does texgenrnn have a checkpoint like feature? LIke itll save its progress mid generation just to be safe? Or, is there a way to improve upon a previously made model?
The indicies_list generation code is way too slow. After about ~30000 texts it almost stops. i7 6700k, 32Gb ram, none of the resources used completely. A simple non-numpy solution generates...
According to the README.md, `textgenrnn` comes with a model that can be trained further. However, as far as I can see, the model that comes with `textgenrnn` is either broken...
I excute the code textgen = textgenrnn() textgen.reset() file="reddit_rarepuppers_politics_2000_context.csv" textgen.train_from_file(file, context=True,new_model=True, num_epochs=10, gen_epochs=5, word_level=True) if i use GPU to run this code, it's OK. but with CPU, it gave me...
Is it possible to train a text gen model for online learning, and update the vocabulary and model parameters continuously one line of text at a time?
