File size: 6,268 Bytes
c407ed8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
160
161
162
163
164
165
from __future__ import annotations

import math

import torch
from torch import nn
from torch.nn import functional as F

from tiny_transformer.config import ModelConfig


class CausalSelfAttention(nn.Module):
    def __init__(self, config: ModelConfig) -> None:
        super().__init__()
        self.n_head = config.n_head
        self.head_dim = config.n_embd // config.n_head
        self.qkv = nn.Linear(config.n_embd, 3 * config.n_embd)
        self.proj = nn.Linear(config.n_embd, config.n_embd)
        self.attn_dropout = nn.Dropout(config.dropout)
        self.resid_dropout = nn.Dropout(config.dropout)
        mask = torch.tril(torch.ones(config.block_size, config.block_size))
        self.register_buffer("causal_mask", mask.view(1, 1, config.block_size, config.block_size))

    def forward(
        self, x: torch.Tensor, return_attention: bool = False
    ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
        batch, seq_len, channels = x.shape
        qkv = self.qkv(x)
        query, key, value = qkv.split(channels, dim=2)

        query = query.view(batch, seq_len, self.n_head, self.head_dim).transpose(1, 2)
        key = key.view(batch, seq_len, self.n_head, self.head_dim).transpose(1, 2)
        value = value.view(batch, seq_len, self.n_head, self.head_dim).transpose(1, 2)

        scores = query @ key.transpose(-2, -1) / math.sqrt(self.head_dim)
        scores = scores.masked_fill(self.causal_mask[:, :, :seq_len, :seq_len] == 0, float("-inf"))
        weights = F.softmax(scores, dim=-1)
        weights = self.attn_dropout(weights)
        out = weights @ value
        out = out.transpose(1, 2).contiguous().view(batch, seq_len, channels)
        out = self.resid_dropout(self.proj(out))
        if return_attention:
            return out, weights
        return out


class FeedForward(nn.Module):
    def __init__(self, config: ModelConfig) -> None:
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(config.n_embd, 4 * config.n_embd),
            nn.GELU(),
            nn.Linear(4 * config.n_embd, config.n_embd),
            nn.Dropout(config.dropout),
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return self.net(x)


class TransformerBlock(nn.Module):
    def __init__(self, config: ModelConfig) -> None:
        super().__init__()
        self.ln_1 = nn.LayerNorm(config.n_embd)
        self.attn = CausalSelfAttention(config)
        self.ln_2 = nn.LayerNorm(config.n_embd)
        self.ffwd = FeedForward(config)

    def forward(
        self, x: torch.Tensor, return_attention: bool = False
    ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
        if return_attention:
            attn_out, weights = self.attn(self.ln_1(x), return_attention=True)
            x = x + attn_out
            x = x + self.ffwd(self.ln_2(x))
            return x, weights

        x = x + self.attn(self.ln_1(x))
        x = x + self.ffwd(self.ln_2(x))
        return x


class TinyTransformer(nn.Module):
    def __init__(self, config: ModelConfig) -> None:
        super().__init__()
        self.config = config
        self.token_embedding = nn.Embedding(config.vocab_size, config.n_embd)
        self.position_embedding = nn.Embedding(config.block_size, config.n_embd)
        self.dropout = nn.Dropout(config.dropout)
        self.blocks = nn.ModuleList([TransformerBlock(config) for _ in range(config.n_layer)])
        self.ln_f = nn.LayerNorm(config.n_embd)
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size)
        self.apply(self._init_weights)

    def _init_weights(self, module: nn.Module) -> None:
        if isinstance(module, nn.Linear):
            nn.init.normal_(module.weight, mean=0.0, std=0.02)
            if module.bias is not None:
                nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            nn.init.normal_(module.weight, mean=0.0, std=0.02)

    def forward(
        self, idx: torch.Tensor, targets: torch.Tensor | None = None
    ) -> tuple[torch.Tensor, torch.Tensor | None]:
        logits, loss, _ = self._forward(idx, targets, capture_attention=False)
        return logits, loss

    def _forward(
        self,
        idx: torch.Tensor,
        targets: torch.Tensor | None = None,
        capture_attention: bool = False,
    ) -> tuple[torch.Tensor, torch.Tensor | None, list[torch.Tensor]]:
        batch, seq_len = idx.shape
        if seq_len > self.config.block_size:
            raise ValueError("Sequence length exceeds block_size")

        attentions: list[torch.Tensor] = []
        positions = torch.arange(seq_len, device=idx.device)
        x = self.token_embedding(idx) + self.position_embedding(positions)
        x = self.dropout(x)
        for block in self.blocks:
            if capture_attention:
                x, weights = block(x, return_attention=True)
                attentions.append(weights)
            else:
                x = block(x)
        x = self.ln_f(x)
        logits = self.lm_head(x)

        loss = None
        if targets is not None:
            loss = F.cross_entropy(logits.view(batch * seq_len, -1), targets.view(batch * seq_len))
        return logits, loss, attentions

    @torch.no_grad()
    def attention_maps(self, idx: torch.Tensor) -> list[torch.Tensor]:
        self.eval()
        _, _, attentions = self._forward(idx, capture_attention=True)
        return attentions

    @torch.no_grad()
    def generate(
        self,
        idx: torch.Tensor,
        max_new_tokens: int,
        temperature: float = 1.0,
        top_k: int | None = None,
    ) -> torch.Tensor:
        if temperature <= 0:
            raise ValueError("temperature must be positive")

        for _ in range(max_new_tokens):
            idx_cond = idx[:, -self.config.block_size :]
            logits, _ = self(idx_cond)
            logits = logits[:, -1, :] / temperature
            if top_k is not None:
                values, _ = torch.topk(logits, min(top_k, logits.size(-1)))
                logits[logits < values[:, [-1]]] = -float("inf")
            probs = F.softmax(logits, dim=-1)
            next_idx = torch.multinomial(probs, num_samples=1)
            idx = torch.cat((idx, next_idx), dim=1)
        return idx