192 lines
6.5 KiB
Python
192 lines
6.5 KiB
Python
|
import torch
|
||
|
import torch.nn as nn
|
||
|
from torch.nn import functional as F
|
||
|
import os
|
||
|
# hyperparameters
|
||
|
batch_size = 32 # how many independent sequences will we process in parallel?
|
||
|
block_size = 5 # what is the maximum context length for predictions?
|
||
|
max_iters = 2500
|
||
|
eval_interval = 50
|
||
|
learning_rate = 1e-4
|
||
|
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
||
|
eval_iters = 200
|
||
|
n_embd = 10
|
||
|
n_head = 5
|
||
|
n_layer = 5
|
||
|
dropout = 0.2
|
||
|
# ------------
|
||
|
|
||
|
torch.manual_seed(1337)
|
||
|
|
||
|
B = 1
|
||
|
T = 40
|
||
|
C = 10
|
||
|
|
||
|
|
||
|
vocab_size = 10
|
||
|
|
||
|
class Head(nn.Module):
|
||
|
""" one head of self-attention """
|
||
|
|
||
|
def __init__(self, head_size, block_size):
|
||
|
super().__init__()
|
||
|
self.key = nn.Linear(n_embd, head_size, bias=False)
|
||
|
self.query = nn.Linear(n_embd, head_size, bias=False)
|
||
|
self.value = nn.Linear(n_embd, head_size, bias=False)
|
||
|
self.register_buffer('tril', torch.tril(torch.ones(block_size, block_size)))
|
||
|
|
||
|
self.dropout = nn.Dropout(dropout)
|
||
|
|
||
|
def forward(self, x):
|
||
|
B,T,C = x.shape
|
||
|
k = self.key(x) # (B,T,C)
|
||
|
q = self.query(x) # (B,T,C)
|
||
|
# compute attention scores ("affinities")
|
||
|
wei = q @ k.transpose(-2,-1) * C**-0.5 # (B, T, C) @ (B, C, T) -> (B, T, T)
|
||
|
wei = wei.masked_fill(self.tril[:T, :T] == 0, float('-inf')) # (B, T, T)
|
||
|
wei = F.softmax(wei, dim=-1) # (B, T, T)
|
||
|
wei = self.dropout(wei)
|
||
|
# perform the weighted aggregation of the values
|
||
|
v = self.value(x) # (B,T,C)
|
||
|
out = wei @ v # (B, T, T) @ (B, T, C) -> (B, T, C)
|
||
|
return out
|
||
|
|
||
|
class MultiHeadAttention(nn.Module):
|
||
|
""" multiple heads of self-attention in parallel """
|
||
|
|
||
|
def __init__(self, num_heads, head_size, block_size):
|
||
|
super().__init__()
|
||
|
self.heads = nn.ModuleList([Head(head_size, block_size) for _ in range(num_heads)])
|
||
|
self.proj = nn.Linear(n_embd, n_embd)
|
||
|
self.dropout = nn.Dropout(dropout)
|
||
|
|
||
|
def forward(self, x):
|
||
|
out = torch.cat([h(x) for h in self.heads], dim=-1)
|
||
|
out = self.dropout(self.proj(out))
|
||
|
return out
|
||
|
|
||
|
class FeedFoward(nn.Module):
|
||
|
""" a simple linear layer followed by a non-linearity """
|
||
|
|
||
|
def __init__(self, n_embd):
|
||
|
super().__init__()
|
||
|
self.net = nn.Sequential(
|
||
|
nn.Linear(n_embd, 4 * n_embd),
|
||
|
nn.ReLU(),
|
||
|
nn.Linear(4 * n_embd, n_embd),
|
||
|
nn.Dropout(dropout),
|
||
|
)
|
||
|
|
||
|
def forward(self, x):
|
||
|
return self.net(x)
|
||
|
|
||
|
class Block(nn.Module):
|
||
|
""" Transformer block: communication followed by computation """
|
||
|
|
||
|
def __init__(self, n_embd, n_head, block_size):
|
||
|
# n_embd: embedding dimension, n_head: the number of heads we'd like
|
||
|
super().__init__()
|
||
|
head_size = n_embd // n_head
|
||
|
self.sa = MultiHeadAttention(n_head, head_size, block_size)
|
||
|
self.ffwd = FeedFoward(n_embd)
|
||
|
self.ln1 = nn.LayerNorm(n_embd)
|
||
|
self.ln2 = nn.LayerNorm(n_embd)
|
||
|
|
||
|
def forward(self, x):
|
||
|
x = x + self.sa(self.ln1(x))
|
||
|
x = x + self.ffwd(self.ln2(x))
|
||
|
return x
|
||
|
|
||
|
|
||
|
class InterBlock(nn.Module):
|
||
|
""" Transformer block: communication followed by computation """
|
||
|
|
||
|
def __init__(self, n_embd, n_head, block_size):
|
||
|
# n_embd: embedding dimension, n_head: the number of heads we'd like
|
||
|
super().__init__()
|
||
|
head_size = n_embd // n_head
|
||
|
self.sa = MultiHeadAttention(n_head, head_size, block_size)
|
||
|
self.ffwd = FeedFoward(n_embd)
|
||
|
self.ln1 = nn.LayerNorm(n_embd)
|
||
|
self.ln2 = nn.LayerNorm(n_embd)
|
||
|
|
||
|
def forward(self, x):
|
||
|
x = x.view(B*T, C, 1)
|
||
|
x = x + self.sa(self.ln1(x))
|
||
|
x = x + self.ffwd(self.ln2(x))
|
||
|
x = x.view(B, T, C)
|
||
|
return x
|
||
|
|
||
|
# super simple bigram model
|
||
|
class BigramLanguageModel(nn.Module):
|
||
|
|
||
|
def __init__(self):
|
||
|
super().__init__()
|
||
|
# each token directly reads off the logits for the next token from a lookup table
|
||
|
self.token_embedding_table = nn.Embedding(vocab_size, n_embd)
|
||
|
self.position_embedding_table = nn.Embedding(block_size, n_embd)
|
||
|
self.pos_emb_inter = nn.Embedding(10, 1)
|
||
|
self.blocks = nn.Sequential(*[Block(n_embd, n_head=n_head, block_size=block_size) for _ in range(n_layer)])
|
||
|
self.interBlocks = nn.Sequential(*[InterBlock(1, n_head=1, block_size=1) for _ in range(n_layer)])
|
||
|
self.l1 = nn.Linear(n_embd, 1000)
|
||
|
self.l2 = nn.Linear(1000, 1000)
|
||
|
self.l3 = nn.Linear(1000, n_embd)
|
||
|
self.ln_f = nn.LayerNorm(n_embd) # final layer norm
|
||
|
self.lm_head = nn.Linear(n_embd, vocab_size)
|
||
|
self.lm_head2 = nn.Linear(n_embd, vocab_size)
|
||
|
self.lm_head3 = nn.Linear(n_embd, vocab_size)
|
||
|
self.lm_head4 = nn.Linear(n_embd, vocab_size)
|
||
|
self.tanh = nn.Tanh()
|
||
|
self.softmax = nn.Softmax(dim=1)
|
||
|
|
||
|
def forward(self, idx, targets=None):
|
||
|
B, T, C = idx.shape
|
||
|
# idx and targets are both (B,T) tensor of integers
|
||
|
# tok_emb = self.token_embedding_table(idx) # (B,T,C)
|
||
|
pos_emb = self.position_embedding_table(torch.arange(T, device=device)) # (T,C)
|
||
|
#pos_emb_inter = self.pos_emb_inter(torch.arange(C, device=device))
|
||
|
|
||
|
x = idx.view(B,T,C)
|
||
|
#x = x + pos_emb_inter
|
||
|
x = idx + pos_emb # (B,T,C)
|
||
|
x = self.blocks(x) # (B,T,C)
|
||
|
#x = self.interBlocks(x) # (B,T,C)
|
||
|
# x = self.l1(x)
|
||
|
# x = self.softmax(x)
|
||
|
# x = self.l2(x)
|
||
|
# x = self.softmax(x)
|
||
|
# x = self.l3(x)
|
||
|
# x = self.softmax(x)
|
||
|
x = self.ln_f(x) # (B,T,C)
|
||
|
logits = self.lm_head(x) # (B,T,vocab_size)
|
||
|
logits = self.softmax(logits)
|
||
|
if targets is None:
|
||
|
loss = None
|
||
|
else:
|
||
|
loss = F.cross_entropy(logits, targets)
|
||
|
|
||
|
return logits, loss
|
||
|
|
||
|
def generate(self, idx, max_new_tokens):
|
||
|
# idx is (B, T) array of indices in the current context
|
||
|
for _ in range(max_new_tokens):
|
||
|
# crop idx to the last block_size tokens
|
||
|
idx_cond = idx[:, -block_size:]
|
||
|
# get the predictions
|
||
|
logits, loss = self(idx_cond)
|
||
|
|
||
|
# focus only on the last time step
|
||
|
logits = logits[:, -1, :] # becomes (B, C)
|
||
|
|
||
|
# B, C = logits.shape
|
||
|
# # apply softmax to get probabilities
|
||
|
# probs = F.softmax(logits, dim=-1) # (B, C)
|
||
|
|
||
|
# # sample from the distribution
|
||
|
# idx_next = torch.multinomial(probs, num_samples=100) # (B, 1)
|
||
|
# append sampled index to the running sequence
|
||
|
|
||
|
logits = logits.view(1, B, C)
|
||
|
idx = torch.cat((idx, logits), dim=1) # (B, T+1)
|
||
|
return idx
|