File size: 5,049 Bytes
3479f48
b999262
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1375e49
3479f48
 
 
 
b999262
 
 
1375e49
b999262
 
 
 
 
3479f48
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
import json
import torch
import torch.nn as nn
from torch.nn import functional as F

device = 'cuda' if torch.cuda.is_available() else 'cpu'


# one head of self-attention using scaled-dot product attention
class Head(nn.Module):
    def __init__(self, n_embed, head_size, context_size, dropout=0.1):
        super().__init__()
        
        self.key = nn.Linear(n_embed, head_size, bias=False)
        self.query = nn.Linear(n_embed, head_size, bias=False)
        self.value = nn.Linear(n_embed, head_size, bias=False)
        self.register_buffer('tril', torch.tril(torch.ones(context_size, context_size)))

        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        B,T,C = x.shape
        k = self.key(x)
        q = self.query(x)
        v = self.value(x)

        tril = torch.tril(torch.ones(T, T, device=device))
        wei = q @ k.transpose(-2, -1) * (C**-0.5)
        wei = wei.masked_fill(tril == 0, float('-inf'))
        wei = F.softmax(wei, dim=-1)
        wei = self.dropout(wei)
        out = wei @ v
        return out


class MultiHeadAttention(nn.Module):
    def __init__(self, n_embed, num_heads, context_size, head_size, dropout):
        super().__init__()

        self.heads = nn.ModuleList([
            Head(n_embed, head_size, context_size)
            for _ in range(num_heads)
        ])
        self.projection = nn.Linear(n_embed, n_embed)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        out = torch.cat([h(x) for h in self.heads], dim=-1)
        out = self.projection(out)
        return self.dropout(out)


# simple feed forward layer
class FeedForward(nn.Module):
    def __init__(self, n_embeds, dropout):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(n_embeds, 4 * n_embeds),
            nn.ReLU(),
            # projection layer
            nn.Linear(4 * n_embeds, n_embeds),
            nn.Dropout(dropout)
        )

    def forward(self, x):
        return self.net(x)


# Transformer block
class Block(nn.Module):
    def __init__(self, n_embeds, n_head, context_size, dropout):
        super().__init__()
        head_size = n_embeds // n_head
        self.sa = MultiHeadAttention(n_embeds, n_head, context_size, head_size, dropout)
        self.ffwd = FeedForward(n_embeds, dropout)
        self.ln1 = nn.LayerNorm(n_embeds)
        self.ln2 = nn.LayerNorm(n_embeds)

    def forward(self, x):
        x = x + self.sa(self.ln1(x))
        x = x + self.ffwd(self.ln2(x))
        return x


# simple bigram model
class DecoderTransformer(nn.Module):
    def __init__(self, vocab_size, n_embed, context_size, n_layer, n_head, dropout):
        super().__init__()

        self.token_embedding_table = nn.Embedding(vocab_size, n_embed)
        self.position_embedding_table = nn.Embedding(context_size, n_embed)
        self.blocks = nn.Sequential(
            *[Block(
                n_embeds=n_embed, 
                n_head=n_head, 
                context_size=context_size, 
                dropout=dropout
            ) for _ in range(n_layer)]
        )
        self.ln_f = nn.LayerNorm(n_embed)
        self.lm_head = nn.Linear(n_embed, vocab_size)

    def forward(self, idx, targets=None):
        B, T = idx.shape
        # idx and targets of size (B,T)
        token_embeds = self.token_embedding_table(idx) # yields (B, T, C)
        pos_embeds = self.position_embedding_table(torch.arange(T, device=device))
        x = token_embeds + pos_embeds
        x = self.ln_f(self.blocks(x))
        logits = self.lm_head(x)

        if targets is None:
            return logits, None

        # reshape elements
        B, T, C = logits.shape
        logits = logits.view(B*T,C)
        targets = targets.view(B*T)
        # compute loss (CE)
        loss = F.cross_entropy(logits, targets)
        return logits, loss

    def generate(self, idx, max_new_tokens=50, context_size=None, temperature=1.0):
        if context_size is None:
            context_size = int(self.position_embedding_table.weight.shape[0])
            print(context_size)

        for _ in range(max_new_tokens):
            idx_cond = idx[:, -context_size:]
            logits, loss = self(idx_cond)
            logits = logits[:,-1,:] / temperature
            probs = F.softmax(logits, dim=-1)
            idx_next = torch.multinomial(probs, num_samples=1)
            idx = torch.cat([idx, idx_next], dim=1)
        return idx


class Tokenizer:
    def __init__(self, vocab):
        self.vocab = vocab
        self.stoi = {ch: idx for idx, ch in enumerate(vocab)}
        self.itos = {idx: ch for idx, ch in enumerate(vocab)}

    def encode(self, s):
        return [self.stoi[c] for c in s]

    def decode(self, i):
        return ''.join([self.itos[x] for x in i])

    @classmethod
    def from_pretrained(cls, path):
        with open(path, 'r') as f:
            vocab = json.load(f)
        return cls(vocab)

    def save_pretrained(self, path):
        with open(path, 'w') as f:
            json.dump(self.vocab, f)