working_gpt.py File Reference

Go to the source code of this file.

Classes
class	working_gpt.Head
	one head of self-attention More...
class	working_gpt.MultiHeadAttention
	multiple heads of self-attention in parallel More...
class	working_gpt.FeedFoward
	a simple linear layer followed by a non-linearity More...
class	working_gpt.Block
	Transformer block: communication followed by computation. More...
class	working_gpt.GPTLanguageModel

Namespaces
namespace	working_gpt

Functions
	working_gpt.get_batch (split)
	working_gpt.estimate_loss ()

Variables
int	working_gpt.batch_size = 64
int	working_gpt.block_size = 256
int	working_gpt.max_iters = 5000
int	working_gpt.eval_interval = 500
int	working_gpt.learning_rate = 3e-4
str	working_gpt.device = "mps"
int	working_gpt.eval_iters = 200
int	working_gpt.n_embd = 384
int	working_gpt.n_head = 6
int	working_gpt.n_layer = 6
float	working_gpt.dropout = 0.2
	working_gpt.encoding
	working_gpt.text = f.read()
	working_gpt.chars = sorted(list(set(text)))
	working_gpt.vocab_size = len(chars)
dict	working_gpt.stoi = {ch: i for i, ch in enumerate(chars)}
dict	working_gpt.itos = {i: ch for i, ch in enumerate(chars)}
	working_gpt.encode
str	working_gpt.decode
	working_gpt.data = torch.tensor(encode(text), dtype=torch.long)
	working_gpt.n = int(0.9 * len(data))
	working_gpt.train_data = data[:n]
	working_gpt.val_data = data[n:]
	working_gpt.model = GPTLanguageModel()
	working_gpt.m = model.to(device)
	working_gpt.optimizer = torch.optim.AdamW(model.parameters(), lr=learning_rate)
	working_gpt.losses = estimate_loss()
	working_gpt.xb
	working_gpt.yb
	working_gpt.logits
	working_gpt.loss
	working_gpt.set_to_none
	working_gpt.context = torch.zeros((1, 1), dtype=torch.long, device=device)