Add Custom Text File Character-Level Training Support

[Adolfo-GM]

Added support for training a character-level GPT model on custom text files (textfile_char dataset).

Training

python train.py config/train_textfile_char.py --device=cpu --compile=False --eval_iters=20 --log_interval=1 --block_size=256 --batch_size=12 --n_layer=6 --n_head=8 --n_embd=128 --max_iters=2000 --lr_decay_iters=2000 --dropout=0.0

Final Results:

Train Loss: 0.0281
Val Loss: 5.3117
Parameters: 1.19M

Sampling

python sample.py --out_dir=out-textfile-char --device=cpu --start="<u> hi" --num_samples=3

Sample Output:

<u> hi> teat's ikire tis as conony es
<a> thereeawives y
<a> lwou touldod ald I he to mar?
<a> gelony mes! dou nonilpih?
<a> he helll! t g alpeo hice? wout blprikire?
<u> it'tthe mesth?
<a> me I any they ane thaly ave blyoing hang memeng! nnin anerey atou cals t fut cer seru blo oto oomio? athind hee.

Added files:

config/
└── train_textfile_char.py       

data/textfile_char/
├── data.txt               
├── prepare.py                  
├── readme.md