Let's build GPT: from scratch, in code, spelled out. —— Karpathy 课程学习笔记

with open('input.txt', 'r', encoding='utf-8') as f:
    text = f.read()                      # 整个数据集就是一个大字符串

chars = sorted(list(set(text)))          # 去重排序得到所有出现的字符
vocab_size = len(chars)                  # 65 个字符（含大小写字母、标点、换行）

stoi = {ch: i for i, ch in enumerate(chars)}   # char -> int
itos = {i: ch for i, ch in enumerate(chars)}   # int -> char
encode = lambda s: [stoi[c] for c in s]        # 字符串 -> 整数列表
decode = lambda l: ''.join([itos[i] for i in l]) # 整数列表 -> 字符串

def get_batch(split):
    data = train_data if split == 'train' else val_data
    ix = torch.randint(len(data) - block_size, (batch_size,))   # 随机起点
    x = torch.stack([data[i:i+block_size] for i in ix])          # 输入：block_size 个字符
    y = torch.stack([data[i+1:i+block_size+1] for i in ix])      # 目标：右移一位
    x, y = x.to(device), y.to(device)
    return x, y

@torch.no_grad()
def estimate_loss():
    out = {}
    model.eval()                          # 评估模式：关闭 dropout 等
    for split in ['train', 'val']:
        losses = torch.zeros(eval_iters)
        for k in range(eval_iters):
            X, Y = get_batch(split)
            logits, loss = model(X, Y)
            losses[k] = loss.item()
        out[split] = losses.mean()        # 多个 batch 平均，损失更平滑
    model.train()
    return out

class BigramLanguageModel(nn.Module):
    def __init__(self, vocab_size):
        super().__init__()
        # 每个 token 直接从一个查找表读出"下一个 token 的 logits"
        self.token_embedding_table = nn.Embedding(vocab_size, vocab_size)

    def forward(self, idx, targets=None):
        logits = self.token_embedding_table(idx)   # (B,T,C)
        if targets is None:
            loss = None
        else:
            B, T, C = logits.shape
            logits = logits.view(B*T, C)           # 展平后算交叉熵
            targets = targets.view(B*T)
            loss = F.cross_entropy(logits, targets)
        return logits, loss

    def generate(self, idx, max_new_tokens):
        for _ in range(max_new_tokens):
            logits, loss = self(idx)               # 前向
            logits = logits[:, -1, :]              # 只看最后时间步
            probs = F.softmax(logits, dim=-1)      # 转概率
            idx_next = torch.multinomial(probs, num_samples=1)  # 采样
            idx = torch.cat((idx, idx_next), dim=1) # 拼回去
        return idx

wei = q @ k.transpose(-2, -1) * k.shape[-1]**-0.5  # (B,T,T) 注意力分数
wei = wei.masked_fill(self.tril[:T, :T] == 0, float('-inf'))  # 因果掩码
wei = F.softmax(wei, dim=-1)                        # 归一化成权重
out = wei @ v                                       # 加权聚合 (B,T,hs)

class MultiHeadAttention(nn.Module):
    def __init__(self, num_heads, head_size):
        super().__init__()
        self.heads = nn.ModuleList([Head(head_size) for _ in range(num_heads)])
        self.proj = nn.Linear(head_size * num_heads, n_embd)   # 输出投影
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        out = torch.cat([h(x) for h in self.heads], dim=-1)    # 并行 head 拼接
        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),   # 升维 4 倍
            nn.ReLU(),
            nn.Linear(4 * n_embd, n_embd),   # 降回原维度
            nn.Dropout(dropout),
        )

class Block(nn.Module):
    """ Transformer block: communication followed by computation """
    def __init__(self, n_embd, n_head):
        head_size = n_embd // n_head
        self.sa = MultiHeadAttention(n_head, head_size)
        self.ffwd = FeedFoward(n_embd)
        self.ln1 = nn.LayerNorm(n_embd)      # 预归一化（pre-norm）
        self.ln2 = nn.LayerNorm(n_embd)

    def forward(self, x):
        x = x + self.sa(self.ln1(x))         # 通信：残差 + 预 LayerNorm
        x = x + self.ffwd(self.ln2(x))       # 计算：残差 + 预 LayerNorm
        return x

class GPTLanguageModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.token_embedding_table = nn.Embedding(vocab_size, n_embd)   # 字符 -> 向量
        self.position_embedding_table = nn.Embedding(block_size, n_embd) # 位置 -> 向量
        self.blocks = nn.Sequential(*[Block(n_embd, n_head=n_head) for _ in range(n_layer)])
        self.ln_f = nn.LayerNorm(n_embd)      # 最终层归一化
        self.lm_head = nn.Linear(n_embd, vocab_size)  # 输出 logits
        self.apply(self._init_weights)        # 权重初始化（视频后补）

    def forward(self, idx, targets=None):
        B, T = idx.shape
        tok_emb = self.token_embedding_table(idx)                    # (B,T,C)
        pos_emb = self.position_embedding_table(torch.arange(T, device=device))  # (T,C)
        x = tok_emb + pos_emb                # 词嵌入 + 位置嵌入
        x = self.blocks(x)
        x = self.ln_f(x)
        logits = self.lm_head(x)             # (B,T,vocab_size)
        # ... loss 计算同 bigram

optimizer = torch.optim.AdamW(model.parameters(), lr=3e-4)
# 5000 步训练，每 500 步打印 train/val loss
# 生成：
context = torch.zeros((1, 1), dtype=torch.long, device=device)
print(decode(m.generate(context, max_new_tokens=500)[0].tolist()))