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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261
| @dataclass class ModelArgs: dim: int = 4096 n_layers: int = 32 n_heads: int = 32 n_kv_heads: Optional[int] = None vocab_size: int = -1 multiple_of: int = 256 ffn_dim_multiplier: Optional[float] = None norm_eps: float = 1e-5
max_batch_size: int = 32 max_seq_len: int = 2048
class RMSNorm(nn.Module): """ 相比 LayerNorm,去掉 mean-centering 和 bias,只保留缩放。 RMSNorm(x) = x / sqrt(mean(x^2) + eps) * weight """ def __init__(self, dim: int, eps: float = 1e-6): super().__init__() self.eps = eps self.weight = nn.Parameter(torch.ones(dim))
def _norm(self, x): return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
def forward(self, x): return self._norm(x.float()).type_as(x) * self.weight
def precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0): """ 预计算 RoPE 的复数频率 e^(i * t * theta_k)。 dim: head_dim end: 最大位置数 返回: [end, dim//2] 的 complex64 张量 """ freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim)) t = torch.arange(end, device=freqs.device) freqs = torch.outer(t, freqs).float() freqs_cis = torch.polar(torch.ones_like(freqs), freqs) return freqs_cis
def reshape_for_broadcast(freqs_cis: torch.Tensor, x: torch.Tensor): """将 freqs_cis 形状调整为可与 x 广播。""" ndim = x.ndim assert 0 <= 1 < ndim assert freqs_cis.shape == (x.shape[1], x.shape[-1]) shape = [d if i == 1 or i == ndim - 1 else 1 for i, d in enumerate(x.shape)] return freqs_cis.view(*shape)
def apply_rotary_emb(xq, xk, freqs_cis): """ 对 query 和 key 应用旋转位置编码。 xq, xk: [B, T, n_heads, head_dim] freqs_cis: [T, head_dim//2] complex 做法: 把 head_dim 视为复数对 (2i, 2i+1),乘以 e^(i*angle) 实现旋转。 """ xq_ = torch.view_as_complex(xq.float().reshape(*xq.shape[:-1], -1, 2)) xk_ = torch.view_as_complex(xk.float().reshape(*xk.shape[:-1], -1, 2)) freqs_cis = reshape_for_broadcast(freqs_cis, xq_) xq_out = torch.view_as_real(xq_ * freqs_cis).flatten(3) xk_out = torch.view_as_real(xk_ * freqs_cis).flatten(3) return xq_out.type_as(xq), xk_out.type_as(xk)
def repeat_kv(x: torch.Tensor, n_rep: int) -> torch.Tensor: """ 当 n_kv_heads < n_heads 时,把 KV heads 复制 n_rep 次以匹配 Q heads。 x: [B, T, n_kv_heads, head_dim] -> [B, T, n_kv_heads * n_rep, head_dim] """ bs, slen, n_kv_heads, head_dim = x.shape if n_rep == 1: return x return ( x[:, :, :, None, :] .expand(bs, slen, n_kv_heads, n_rep, head_dim) .reshape(bs, slen, n_kv_heads * n_rep, head_dim) )
class Attention(nn.Module): def __init__(self, args: ModelArgs): super().__init__() self.n_kv_heads = args.n_heads if args.n_kv_heads is None else args.n_kv_heads self.n_heads = args.n_heads self.n_rep = self.n_heads // self.n_kv_heads self.head_dim = args.dim // args.n_heads
self.wq = nn.Linear(args.dim, args.n_heads * self.head_dim, bias=False) self.wk = nn.Linear(args.dim, self.n_kv_heads * self.head_dim, bias=False) self.wv = nn.Linear(args.dim, self.n_kv_heads * self.head_dim, bias=False) self.wo = nn.Linear(args.n_heads * self.head_dim, args.dim, bias=False)
self.cache_k = torch.zeros( (args.max_batch_size, args.max_seq_len, self.n_kv_heads, self.head_dim) ) self.cache_v = torch.zeros( (args.max_batch_size, args.max_seq_len, self.n_kv_heads, self.head_dim) )
def forward(self, x, start_pos, freqs_cis, mask): bsz, seqlen, _ = x.shape xq, xk, xv = self.wq(x), self.wk(x), self.wv(x)
xq = xq.view(bsz, seqlen, self.n_heads, self.head_dim) xk = xk.view(bsz, seqlen, self.n_kv_heads, self.head_dim) xv = xv.view(bsz, seqlen, self.n_kv_heads, self.head_dim)
xq, xk = apply_rotary_emb(xq, xk, freqs_cis=freqs_cis)
self.cache_k = self.cache_k.to(xq) self.cache_v = self.cache_v.to(xq) self.cache_k[:bsz, start_pos:start_pos + seqlen] = xk self.cache_v[:bsz, start_pos:start_pos + seqlen] = xv
keys = self.cache_k[:bsz, :start_pos + seqlen] values = self.cache_v[:bsz, :start_pos + seqlen]
keys = repeat_kv(keys, self.n_rep) values = repeat_kv(values, self.n_rep)
xq = xq.transpose(1, 2) keys = keys.transpose(1, 2) values = values.transpose(1, 2)
scores = torch.matmul(xq, keys.transpose(2, 3)) / math.sqrt(self.head_dim) if mask is not None: scores = scores + mask scores = F.softmax(scores.float(), dim=-1).type_as(xq)
output = torch.matmul(scores, values) output = output.transpose(1, 2).contiguous().view(bsz, seqlen, -1) return self.wo(output)
class FeedForward(nn.Module): """ SwiGLU: FFN(x) = W2( SiLU(W1 x) * W3 x ) hidden_dim 通常取 4*dim 的 2/3,且向上取整到 multiple_of 的倍数。 """ def __init__(self, dim, hidden_dim, multiple_of, ffn_dim_multiplier=None): super().__init__() hidden_dim = int(2 * hidden_dim / 3) if ffn_dim_multiplier is not None: hidden_dim = int(ffn_dim_multiplier * hidden_dim) hidden_dim = multiple_of * ((hidden_dim + multiple_of - 1) // multiple_of)
self.w1 = nn.Linear(dim, hidden_dim, bias=False) self.w2 = nn.Linear(hidden_dim, dim, bias=False) self.w3 = nn.Linear(dim, hidden_dim, bias=False)
def forward(self, x): return self.w2(F.silu(self.w1(x)) * self.w3(x))
class TransformerBlock(nn.Module): """ Pre-Norm 结构: h = x + Attention(RMSNorm(x)) out = h + FFN(RMSNorm(h)) """ def __init__(self, layer_id, args): super().__init__() self.n_heads = args.n_heads self.dim = args.dim self.head_dim = args.dim // args.n_heads self.attention = Attention(args) self.feed_forward = FeedForward( dim=args.dim, hidden_dim=4 * args.dim, multiple_of=args.multiple_of, ffn_dim_multiplier=args.ffn_dim_multiplier, ) self.layer_id = layer_id self.attention_norm = RMSNorm(args.dim, eps=args.norm_eps) self.ffn_norm = RMSNorm(args.dim, eps=args.norm_eps)
def forward(self, x, start_pos, freqs_cis, mask): h = x + self.attention(self.attention_norm(x), start_pos, freqs_cis, mask) out = h + self.feed_forward(self.ffn_norm(h)) return out
class Transformer(nn.Module): def __init__(self, params: ModelArgs): super().__init__() self.params = params self.vocab_size = params.vocab_size self.n_layers = params.n_layers
self.tok_embeddings = nn.Embedding(params.vocab_size, params.dim)
self.layers = nn.ModuleList() for layer_id in range(params.n_layers): self.layers.append(TransformerBlock(layer_id, params))
self.norm = RMSNorm(params.dim, eps=params.norm_eps) self.output = nn.Linear(params.dim, params.vocab_size, bias=False)
self.freqs_cis = precompute_freqs_cis( self.params.dim // self.params.n_heads, self.params.max_seq_len * 2, )
@torch.inference_mode() def forward(self, tokens, start_pos): """ tokens: [B, T] start_pos: 当前 token 在序列中的起始位置(用于 KV Cache 和 RoPE) """ _bsz, seqlen = tokens.shape h = self.tok_embeddings(tokens) self.freqs_cis = self.freqs_cis.to(h.device) freqs_cis = self.freqs_cis[start_pos:start_pos + seqlen]
mask = None if seqlen > 1: mask = torch.full((seqlen, seqlen), float("-inf"), device=tokens.device) mask = torch.triu(mask, diagonal=1) mask = torch.hstack([ torch.zeros((seqlen, start_pos), device=tokens.device), mask ]).type_as(h)
for layer in self.layers: h = layer(h, start_pos, freqs_cis, mask) h = self.norm(h) output = self.output(h).float() return output
|