|
--- |
|
license: mit |
|
tags: |
|
- natural-language-processing |
|
- code-generation |
|
- torch |
|
- lstm |
|
--- |
|
|
|
This generative text model was trained using [Andrej Karpathy's code](https://github.com/karpathy/char-rnn) on homeworks by Linguistics students' homework assignments for a beginning Python course of HSE University. |
|
|
|
Model was trained with size 512 and 3 layers, dropout 0.5. |
|
|
|
## Usage |
|
|
|
The procedure for installing the required software is described [by Karpathy](https://github.com/karpathy/char-rnn), torch is required, the code is written in lua. Be careful, versions of libraries written many years ago are used. |
|
|
|
```bash |
|
th sample.lua lm_lstm_epoch19.14_0.6292.t7 -length 10000 -temperature 0.5 -primetext 'some text' |
|
``` |
|
|
|
## Train data |
|
|
|
Train corpus consists of joined programms in to one file inclded in this repository as `input.txt` |
|
|
|
|
|
## What for? |
|
|
|
In an era of winning Transformers, ancient RNN models seem archaic. But I see that they still work better than modern architectures with such important categories from the humanities point of view as individual style. |
|
|
|
This model was created just or fun of students at the end of the course. |
|
|
|
|
|
|