File size: 13,103 Bytes
0e28a9a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 |
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers.configuration_utils import PretrainedConfig
class DepthGPTConfig(PretrainedConfig):
def __init__(
self,
block_size: int = 8,
vocab_size: int = 2049, # GPT-2 vocab_size of 50257, padded up to nearest multiple of 64 for efficiency
n_layer: int = 6,
n_head: int = 16,
n_embd: int = 1024,
dropout: float = 0.0,
bias: bool = False, # True: bias in Linears and LayerNorms, like GPT-2. False: a bit better and faster
main_hidden_size = 1536,
pad_token_id = 2048,
use_cmlp = True,
use_rmsnorm = False,
use_swiglu = False
):
"""
{
"block_size": 8,
"vocab_size": 2049,
"n_layer": 6,
"n_head": 16,
"n_embd": 1024,
"dropout": 0.0,
"bias": false,
"main_hidden_size": 1536,
"pad_token_id": 2048,
"use_cmlp": true
}
"""
# super().__init__(**kwargs)
self.block_size = block_size
self.vocab_size = vocab_size
self.n_layer = n_layer
self.n_head = n_head
self.n_embd = n_embd
self.dropout = dropout
self.bias = bias
self.main_hidden_size = main_hidden_size
self.pad_token_id = pad_token_id
self.use_cmlp = use_cmlp
self.use_rmsnorm = use_rmsnorm
self.use_swiglu = use_swiglu
################################################################################################
# GPT style
################################################################################################
class LayerNorm(nn.Module):
""" LayerNorm but with an optional bias. PyTorch doesn't support simply bias=False """
def __init__(self, ndim, bias):
super().__init__()
self.weight = nn.Parameter(torch.ones(ndim))
self.bias = nn.Parameter(torch.zeros(ndim)) if bias else None
def forward(self, input):
return F.layer_norm(input, self.weight.shape, self.weight, self.bias, 1e-5)
class RMSNorm(nn.Module):
def __init__(self, dim: int, eps: float = 1e-6):
super(RMSNorm, self).__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):
output = self._norm(x.float()).type_as(x)
return output * self.weight
class CausalSelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
assert config.n_embd % config.n_head == 0
# key, query, value projections for all heads, but in a batch
self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd, bias=config.bias)
# output projection
self.c_proj = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
# regularization
self.attn_dropout = nn.Dropout(config.dropout)
self.resid_dropout = nn.Dropout(config.dropout)
self.n_head = config.n_head
self.n_embd = config.n_embd
self.dropout = config.dropout
# flash attention make GPU go brrrrr but support is only in PyTorch >= 2.0
self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention')
if not self.flash:
print("WARNING: using slow attention. Flash Attention requires PyTorch >= 2.0")
# causal mask to ensure that attention is only applied to the left in the input sequence
self.register_buffer("bias", torch.tril(torch.ones(config.block_size, config.block_size))
.view(1, 1, config.block_size, config.block_size))
def forward(self, x):
B, T, C = x.size() # batch size, sequence length, embedding dimensionality (n_embd)
# calculate query, key, values for all heads in batch and move head forward to be the batch dim
q, k, v = self.c_attn(x).split(self.n_embd, dim=2)
k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
# causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
if self.flash:
# efficient attention using Flash Attention CUDA kernels
y = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=self.dropout if self.training else 0, is_causal=True)
else:
# manual implementation of attention
att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
att = att.masked_fill(self.bias[:,:,:T,:T] == 0, float('-inf'))
att = F.softmax(att, dim=-1)
att = self.attn_dropout(att)
y = att @ v # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
y = y.transpose(1, 2).contiguous().view(B, T, C) # re-assemble all head outputs side by side
# output projection
y = self.resid_dropout(self.c_proj(y))
return y
class MLP(nn.Module):
def __init__(self, config):
super().__init__()
self.c_fc = nn.Linear(config.n_embd, 4 * config.n_embd, bias=config.bias)
self.gelu = nn.GELU()
self.c_proj = nn.Linear(4 * config.n_embd, config.n_embd, bias=config.bias)
self.dropout = nn.Dropout(config.dropout)
def forward(self, x):
x = self.c_fc(x)
x = self.gelu(x)
x = self.c_proj(x)
x = self.dropout(x)
return x
class MLP_swiglu(nn.Module):
def __init__(self, config):
super().__init__()
self.intermediate_size = int(8 * config.n_embd / 3)
self.gate_proj = nn.Linear(config.n_embd, self.intermediate_size, bias=config.bias)
self.up_proj = nn.Linear(config.n_embd, self.intermediate_size, bias=config.bias)
self.down_proj = nn.Linear(self.intermediate_size, config.n_embd, bias=config.bias)
self.act_fn = F.silu
self.dropout = nn.Dropout(config.dropout)
def forward(self, x):
x = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
x = self.dropout(x)
return x
class Block(nn.Module):
def __init__(self, config):
super().__init__()
self.ln_1 = RMSNorm(config.n_embd) if config.use_rmsnorm else LayerNorm(config.n_embd, bias=config.bias)
self.attn = CausalSelfAttention(config)
self.ln_2 = RMSNorm(config.n_embd) if config.use_rmsnorm else LayerNorm(config.n_embd, bias=config.bias)
mlp_cls = MLP_swiglu if config.use_swiglu else MLP
self.mlp = mlp_cls(config)
def forward(self, x):
x = x + self.attn(self.ln_1(x))
x = x + self.mlp(self.ln_2(x))
return x
class BlockCMLP(nn.Module):
def __init__(self, config):
super().__init__()
self.channel_size = config.block_size
self.ln_1 = RMSNorm(config.n_embd) if config.use_rmsnorm else LayerNorm(config.n_embd, bias=config.bias)
self.attn = CausalSelfAttention(config)
self.ln_2 = RMSNorm(config.n_embd) if config.use_rmsnorm else LayerNorm(config.n_embd, bias=config.bias)
mlp_cls = MLP_swiglu if config.use_swiglu else MLP
self.mlps = nn.ModuleList([mlp_cls(config) for _ in range(self.channel_size)])
assert self.channel_size == 8, f"DEBUG, self.channel_size={self.channel_size} != 8"
def forward(self, x):
_, channel_size, _ = x.shape
# assert channel_size == self.channel_size
x = x + self.attn(self.ln_1(x))
xl = self.ln_2(x)
x = x + torch.cat(
[self.mlps[c](xl[:, c:c+1, :]) for c in range(self.channel_size)],
dim=1
)
return x
class DepthGPT(nn.Module):
def __init__(self, config):
super().__init__()
assert config.vocab_size is not None
assert config.block_size is not None
self.config = config
self.num_channel = config.block_size
self.linear_in = nn.Linear(config.main_hidden_size, config.n_embd * config.block_size, bias=False)
block_cls = BlockCMLP if config.use_cmlp else Block
self.transformer = nn.ModuleDict(dict(
wtes = nn.ModuleList([nn.Embedding(config.vocab_size, config.n_embd) for _ in range(self.num_channel)]),
wpe = nn.Embedding(self.num_channel, config.n_embd),
drop = nn.Dropout(config.dropout),
h = nn.ModuleList([block_cls(config) for _ in range(config.n_layer)]),
ln_f = RMSNorm(config.n_embd) if config.use_rmsnorm else LayerNorm(config.n_embd, bias=config.bias),
))
self.lm_heads = nn.ModuleList([nn.Linear(config.n_embd, config.vocab_size, bias=False) for _ in range(self.num_channel)])
# with weight tying when using torch.compile() some warnings get generated:
# "UserWarning: functional_call was passed multiple values for tied weights.
# This behavior is deprecated and will be an error in future versions"
# not 100% sure what this is, so far seems to be harmless. TODO investigate
# self.transformer.wte.weight = self.lm_head.weight # https://paperswithcode.com/method/weight-tying
# init all weights
self.apply(self._init_weights)
# apply special scaled init to the residual projections, per GPT-2 paper
for pn, p in self.named_parameters():
if pn.endswith('c_proj.weight'):
torch.nn.init.normal_(p, mean=0.0, std=0.02/math.sqrt(2 * config.n_layer))
# report number of parameters
print("number of parameters: %.2fM" % (self.get_num_params()/1e6,))
def get_num_params(self, non_embedding=False):
"""
Return the number of parameters in the model.
For non-embedding count (default), the position embeddings get subtracted.
The token embeddings would too, except due to the parameter sharing these
params are actually used as weights in the final layer, so we include them.
"""
n_params = sum(p.numel() for p in self.parameters())
if non_embedding:
n_params -= self.transformer.wpe.weight.numel()
return n_params
def _init_weights(self, module):
if isinstance(module, nn.Linear):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
if module.bias is not None:
torch.nn.init.zeros_(module.bias)
elif isinstance(module, nn.Embedding):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
def forward(self,
main_hidden_states, # [seq, main_dim]
audio_token_ids # [seq, 7]
):
assert main_hidden_states.shape[0] == audio_token_ids.shape[0]
in_audio_token_num = audio_token_ids.shape[-1]
device = audio_token_ids.device
audio_token_ids = F.pad(audio_token_ids, (1, 0), value=self.config.pad_token_id)
x = torch.stack(
[self.transformer.wtes[c](audio_token_ids[:, c]) for c in range(in_audio_token_num + 1)]
).transpose(0, 1) # [seq, in_audio_token_num]
x += self.transformer.wpe(
torch.arange(0, in_audio_token_num + 1, dtype=torch.long, device=device)
).unsqueeze(0) # position embeddings of shape (1, 8, depth_dim)
main_hidden = self.linear_in(main_hidden_states).view(main_hidden_states.shape[0], self.config.block_size, -1)[:, :in_audio_token_num+1, :]
x += main_hidden
x = self.transformer.drop(x)
for block in self.transformer.h:
x = block(x)
# [seq, 8, hidden]
x = self.transformer.ln_f(x)
# [seq, 8, hidden] (linear)-> [8, seq, vocab]
x = torch.stack([self.lm_heads[c](x[:, c, :]) for c in range(x.shape[1])])
# [8, seq, vocab] -> [seq, 8, vocab]
x = x.transpose(0,1)
return x
def _initialize_weights(self, module):
if isinstance(module, nn.Linear):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
if module.bias is not None:
torch.nn.init.zeros_(module.bias)
elif isinstance(module, nn.Embedding):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
if __name__ == "__main__":
config = {
"bias": False,
"dropout": 0.0,
"n_embd": 1024,
"n_head": 16,
"n_layer": 6,
"use_cmlp": True,
"use_rmsnorm": True,
"use_swiglu": True,
"main_hidden_size": 4096
}
model_config = DepthGPTConfig(**config)
model = DepthGPT(config=model_config)
main_hidden_states = torch.rand((1, 4096))
decoded_audio_tokens = torch.empty((1, 0), dtype=torch.long, device=main_hidden_states.device)
outputs = model(main_hidden_states, decoded_audio_tokens)
|