GPT / config.py
HemaAM's picture
Initial commit of application files
a4d575b
raw
history blame contribute delete
380 Bytes
import torch
# hyperparameters
batch_size = 64 # how many independent sequences will we process in parallel?
block_size = 256 # what is the maximum context length for predictions?
max_iters = 5000
eval_interval = 500
learning_rate = 3e-4
device = 'cuda' if torch.cuda.is_available() else 'cpu'
eval_iters = 200
n_embd = 384
n_head = 6
n_layer = 6
dropout = 0.2
# ------------