mygpt.py

   1 #!/usr/bin/env python
   2
   3 # Any copyright is dedicated to the Public Domain.
   4 # https://creativecommons.org/publicdomain/zero/1.0/
   5
   6 # Written by Francois Fleuret <francois@fleuret.org>
   7
   8 # This is an implementation from scratch of a "GPT", that is a model
   9 # composed of several causal self-attention blocks. It is equipped
  10 # with a caching mechanism for keys and values to avoid a O(N^3) cost
  11 # for auto-regression.
  12
  13 # This implementation is equipped with RNN layers to replace the MHA
  14
  15 import math, warnings
  16
  17 import torch, einops
  18
  19 from torch import nn
  20 from torch.nn import functional as F
  21
  22 import ffutils
  23
  24 # import memload
  25
  26 ######################################################################
  27
  28 # A BracketedSequence is a BxTx... tensor with a first and a nb time
  29 # steps to compute.
  30
  31 # Modules able to process it expect that they will have to process a
  32 # first bracket starting at t=0, followed by a succession of brackets
  33 # that move forward in time, do not overlap, and cover the axis T with
  34 # no holes.
  35 #
  36 # Although it is more general, for a classical prompt-conditioned
  37 # auto-regressive process it will be a first bracket starting at 0 and
  38 # of arbitrary length for the "prompt", followed by brackets of length
  39 # 1 for the successive tokens.
  40 #
  41 # Modules able to process brackets may implement a cache that is
  42 # resetted when init_cache is True
  43
  44
  45 class BracketedSequence:
  46     def __init__(self, x, first=None, nb=None, init_cache=None):
  47         self.x = x
  48         assert (first is None and nb is None and init_cache is None) or (
  49             first is not None and nb is not None and init_cache is not None
  50         )
  51
  52         self.first = 0 if first is None else first
  53         self.nb = x.size(1) if nb is None else nb
  54         self.init_cache = True if init_cache is None else init_cache
  55
  56     def slice(self):
  57         return self.x[:, self.first : self.first + self.nb]
  58
  59     def complete(self):
  60         return self.first == 0 and self.nb == self.x.size(1)
  61
  62
  63 ######################################################################
  64
  65
  66 class CacheWrapper(nn.Module):
  67     def __init__(self, *f):
  68         super().__init__()
  69         self.f = f[0] if len(f) == 1 else nn.Sequential(*f)
  70
  71     def forward(self, bs):
  72         if bs.init_cache:
  73             y = self.f(bs.slice())
  74             self.cache_y = y.new(*((y.size(0), bs.x.size(1)) + y.size()[2:]))
  75             self.cache_y[:, bs.first : bs.first + bs.nb] = y
  76         else:
  77             assert tuple(bs.x.size()[:2]) == tuple(self.cache_y.size()[:2])
  78             assert bs.first + bs.nb <= self.cache_y.size(1)
  79             self.cache_y[:, bs.first : bs.first + bs.nb] = self.f(bs.slice())
  80
  81         return BracketedSequence(self.cache_y, bs.first, bs.nb, bs.init_cache)
  82
  83
  84 ##############################
  85
  86
  87 class WithResidual(nn.Module):
  88     def __init__(self, *f):
  89         super().__init__()
  90         self.f = f[0] if len(f) == 1 else nn.Sequential(*f)
  91
  92     def forward(self, bs):
  93         return BracketedSequence(bs.x + self.f(bs).x, bs.first, bs.nb, bs.init_cache)
  94
  95
  96 ##############################
  97
  98
  99 class AddPositionalEncoding(nn.Module):
 100     def __init__(self, len_max):
 101         super().__init__()
 102         self.len_max = len_max
 103
 104     # [Vaswani et al 2018] PE_{t,2i} = sin(t/(L^{2i/D})), PE_{t,2i+1} = cos(t/(L^{2i/D}))
 105
 106     def forward(self, bs):
 107         if bs.init_cache:
 108             t = torch.arange(bs.x.size(1), dtype=bs.x.dtype, device=bs.x.device)[
 109                 :, None
 110             ]
 111             j = torch.arange(bs.x.size(2), dtype=bs.x.dtype, device=bs.x.device)[
 112                 None, :
 113             ]
 114             k = j % 2
 115             self.pe = torch.sin(
 116                 t / (self.len_max ** ((j - k) / bs.x.size(2))) + math.pi / 2 * k
 117             )
 118             self.cache_y = bs.x.new(bs.x.size())
 119
 120         self.cache_y[:, bs.first : bs.first + bs.nb] = (
 121             bs.slice() + self.pe[bs.first : bs.first + bs.nb]
 122         )
 123
 124         return BracketedSequence(self.cache_y, bs.first, bs.nb, bs.init_cache)
 125
 126
 127 import pscan
 128
 129 # X is /.../xTxD   A is /.../xT   Y_init is /.../xD
 130
 131
 132 def pscan_dim(A, X, Y_init, dim=-2):
 133     s = X.size()
 134     a, T, b = s[:dim].numel(), s[dim], s[dim + 1 :].numel()
 135
 136     A = A.reshape(a, T, *s[dim + 1 : -1])
 137     X = X.reshape(a, T, *s[dim + 1 : -1], -1)
 138
 139     if Y_init is None:
 140         Y_init = X.new_zeros(a, *s[dim + 1 : -1], X.size(-1))
 141     else:
 142         Y_init = Y_init.reshape(a, *s[dim + 1 : -1], -1)
 143
 144     Y = pscan.pscan(A, X, Y_init).reshape(s)
 145
 146     return Y
 147
 148
 149 def pscan_rgrad(grad_Y, A, X, Y_init, dim=-2, eps=1e-2):
 150     with torch.no_grad():
 151         s_A, s_X = 0, 0
 152         for t in range(X.size(dim) - 1, 0, -1):
 153             delta = (grad_Y[t] - s_A) / A[t].grad
 154             s_A += A[t].grad * delta
 155             A[t].grad = delta
 156             delta = (grad_Y[t] - s_X) / X[t].grad
 157             s_X += X[t].grad * delta
 158             X[t].grad = delta
 159
 160
 161 def pscan_shape(A, X, Y_init):
 162     s = X.size()
 163     A = A.reshape(-1, s[-2])
 164     X = X.reshape(-1, s[-2], s[-1])
 165
 166     if Y_init is None:
 167         Y_init = X.new_zeros(X.size(0), s[-1])
 168     else:
 169         Y_init = Y_init.reshape(-1, s[-1])
 170
 171     Y = pscan.pscan(A, X, Y_init).reshape(s)
 172
 173     return Y
 174
 175
 176 def nsum_shape(X, Y_init):
 177     s = X.size()
 178     X = X.reshape(-1, s[-2], s[-1])  # ntd
 179
 180     Y = 0 if Y_init is None else Y_init.reshape(-1, s[-1])
 181     result = []
 182
 183     for k in range(X.size(1)):
 184         Y = Y + X[:, k]
 185         Y = Y / Y.norm(dim=-1, keepdim=True).clamp(min=1)
 186         result.append(Y)
 187
 188     return torch.cat(result, dim=1).reshape(s)
 189
 190
 191 ##############################
 192
 193
 194 class DumbRec(nn.Module):
 195     def __init__(
 196         self,
 197         dim_model,
 198         dim_qk,
 199         dim_v,
 200         nb_heads,
 201         nb_lines,
 202         attention_dropout=0.0,
 203         len_max=1e5,
 204         logger=print,
 205         **kwargs,
 206     ):
 207         super().__init__()
 208
 209         def randw(*d):
 210             return nn.Parameter(torch.randn(*d) / math.sqrt(d[-1]))
 211
 212         self.nb_lines = nb_lines
 213         self.attention_dropout = attention_dropout
 214
 215         self.k_star = randw(nb_lines, dim_qk)
 216
 217         self.w_qw = randw(nb_heads, dim_qk, dim_model)
 218         self.w_qr = randw(nb_heads, dim_qk, dim_model)
 219         # self.w_k = randw(nb_heads, dim_qk, dim_model)
 220         self.w_v = randw(nb_heads, dim_v, dim_model)
 221         self.w_o = randw(dim_v * nb_heads, dim_model)
 222
 223     def reset_inner_loss(self):
 224         self.acc_attention = 0
 225         self.acc_nb = 0
 226
 227     def get_inner_loss(self):
 228         warnings.warn("l2 regularization", RuntimeWarning)
 229         return (self.acc_attention / self.acc_nb).pow(2).sum()
 230         # return torch.tensor([0], device=self.w_qw.device)
 231
 232     def forward(self, bs):
 233         x_q, t0, t1 = bs.x, bs.first, bs.first + bs.nb
 234
 235         if bs.init_cache:
 236             self.rec_v = x_q.new_zeros(
 237                 x_q.size(0), self.nb_lines, x_q.size(1), self.w_v.size(1)
 238             )
 239             # self.rec_k = x_q.new_zeros(
 240             # x_q.size(0), self.nb_lines, x_q.size(1), self.w_k.size(1)
 241             # )
 242             self.cache_y = x_q.new_zeros(x_q.size(0), x_q.size(1), self.w_o.size(1))
 243
 244         ######################################################################
 245         # Prepare the keys
 246
 247         k_star = self.k_star[:, None, :].expand(-1, t1 - t0, -1)
 248
 249         warnings.warn("rotating key barrel", RuntimeWarning)
 250         k_star = self.k_star[:, None, :].expand(-1, x_q.size(1), -1)
 251         t_barrel = torch.arange(t0, t1, device=k_star.device)
 252         t_barrel = t_barrel[None, :].expand(k_star.size(0), t1 - t0)
 253         l_barrel = (
 254             torch.arange(k_star.size(0), device=k_star.device)[:, None] + t_barrel
 255         ) % k_star.size(0)
 256         k_star = k_star[l_barrel, t_barrel]
 257
 258         ######################################################################
 259         # Compute the recurrent state
 260
 261         qw = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_qw)
 262
 263         v = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_v)
 264         # k = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_k)
 265
 266         aw = torch.einsum(
 267             "nhtd,ltd->nhlt",
 268             qw,
 269             k_star,
 270         ) / math.sqrt(self.w_qw.size(1))
 271
 272         aw = aw.softmax(dim=2)  # nhlt
 273
 274         if self.train:
 275             self.acc_attention += aw.sum(dim=(0, 1, 3))
 276             self.acc_nb += aw.size(0) * aw.size(1) * aw.size(3)
 277
 278         aw = F.dropout(aw, self.attention_dropout, self.training)
 279
 280         A = 1 - aw.sum(dim=1)  # nlt
 281
 282         V = torch.einsum("nhlt,nhtd->nltd", aw, v).contiguous()
 283         # K = torch.einsum("nhlt,nhtd->nltd", aw, k).contiguous()
 284
 285         if t0 == 0:
 286             V0 = None
 287             # K0 = None
 288         else:
 289             V0 = self.rec_v[:, :, t0 - 1]
 290             # K0 = self.rec_k[:, :, t0 - 1]
 291
 292         self.rec_v[:, :, t0:t1] = pscan_shape(A, V, V0)
 293         # self.rec_k[:, :, t0:t1] = pscan_shape(A, K, K0)
 294
 295         ######################################################################
 296         # compute the readout
 297
 298         qr = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_qr)
 299
 300         ar = torch.einsum(
 301             "nhtd,ld->nhlt",
 302             qr,
 303             # self.rec_k[:, :, t0:t1],
 304             self.k_star,
 305         ) / math.sqrt(self.w_qr.size(1))
 306
 307         ar = ar.softmax(dim=2)  # nhlt
 308
 309         ar = F.dropout(ar, self.attention_dropout, self.training)
 310
 311         y = torch.einsum(
 312             "nhlt,nltd->nthd",
 313             ar,
 314             self.rec_v[:, :, t0:t1],
 315         ).flatten(2)
 316
 317         self.cache_y[:, t0:t1] = y @ self.w_o
 318
 319         return BracketedSequence(self.cache_y, t0, t1 - t0, bs.init_cache)
 320
 321
 322 ##############################
 323
 324
 325 class KVRec(nn.Module):
 326     def __init__(
 327         self,
 328         dim_model,
 329         dim_qk,
 330         dim_v,
 331         nb_heads,
 332         nb_lines,
 333         attention_dropout=0.0,
 334         len_max=1e5,
 335         logger=print,
 336         **kwargs,
 337     ):
 338         super().__init__()
 339
 340         def randw(*d):
 341             return nn.Parameter(torch.randn(*d) / math.sqrt(d[-1]))
 342
 343         self.nb_lines = nb_lines
 344         self.attention_dropout = attention_dropout
 345
 346         self.k_star = randw(nb_lines, dim_qk)
 347
 348         self.w_qw = randw(nb_heads, dim_qk, dim_model)
 349         self.w_qr = randw(nb_heads, dim_qk, dim_model)
 350         self.w_k = randw(nb_heads, dim_qk, dim_model)
 351         self.w_v = randw(nb_heads, dim_v, dim_model)
 352         self.w_o = randw(dim_v * nb_heads, dim_model)
 353
 354     def reset_inner_loss(self):
 355         self.acc_attention = 0
 356         self.acc_nb = 0
 357
 358     def get_inner_loss(self):
 359         warnings.warn("l2 regularization", RuntimeWarning)
 360         return (self.acc_attention / self.acc_nb).pow(2).sum()
 361         # return torch.tensor([0], device=self.w_qw.device)
 362         # warnings.warn("side regularization", RuntimeWarning)
 363         # return (
 364         # (0.5 / self.nb_lines - self.acc_attention / self.acc_nb).clamp(min=0).sum()
 365         # )
 366         # return torch.tensor([0], device=self.w_qw.device)
 367
 368     def forward(self, bs):
 369         x_q, t0, t1 = bs.x, bs.first, bs.first + bs.nb
 370
 371         if bs.init_cache:
 372             self.rec_v = x_q.new_zeros(
 373                 x_q.size(0), self.nb_lines, x_q.size(1), self.w_v.size(1)
 374             )
 375             self.rec_k = x_q.new_zeros(
 376                 x_q.size(0), self.nb_lines, x_q.size(1), self.w_k.size(1)
 377             )
 378             self.cache_y = x_q.new_zeros(x_q.size(0), x_q.size(1), self.w_o.size(1))
 379
 380         ######################################################################
 381         # Prepare the keys
 382
 383         k_star = self.k_star[:, None, :].expand(-1, t1 - t0, -1)
 384
 385         warnings.warn("rotating key barrel", RuntimeWarning)
 386         k_star = self.k_star[:, None, :].expand(-1, x_q.size(1), -1)
 387         t_barrel = torch.arange(t0, t1, device=k_star.device)
 388         t_barrel = t_barrel[None, :].expand(k_star.size(0), t1 - t0)
 389         l_barrel = (
 390             torch.arange(k_star.size(0), device=k_star.device)[:, None] + t_barrel
 391         ) % k_star.size(0)
 392         k_star = k_star[l_barrel, t_barrel]
 393
 394         ######################################################################
 395         # Compute the recurrent state
 396
 397         qw = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_qw)
 398
 399         v = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_v)
 400         k = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_k)
 401
 402         aw = torch.einsum(
 403             "nhtd,ltd->nhlt",
 404             qw,
 405             k_star,
 406         ) / math.sqrt(self.w_qw.size(1))
 407
 408         aw = aw.softmax(dim=2)  # nhlt
 409
 410         if self.train:
 411             # We want all the memory lines to be used similarly
 412             self.acc_attention += aw.sum(dim=(0, 1, 3))  # Sum accross NxHx_xT
 413             self.acc_nb += aw.size(0) * aw.size(1) * aw.size(3)
 414
 415         aw = F.dropout(aw, self.attention_dropout, self.training)
 416
 417         A = 1 - aw.sum(dim=1)  # nlt
 418
 419         V = torch.einsum("nhlt,nhtd->nltd", aw, v).contiguous()
 420         K = torch.einsum("nhlt,nhtd->nltd", aw, k).contiguous()
 421
 422         if t0 == 0:
 423             V0 = None
 424             K0 = None
 425         else:
 426             V0 = self.rec_v[:, :, t0 - 1]
 427             K0 = self.rec_k[:, :, t0 - 1]
 428
 429         self.rec_v[:, :, t0:t1] = pscan_shape(A, V, V0)
 430         self.rec_k[:, :, t0:t1] = pscan_shape(A, K, K0)
 431
 432         ######################################################################
 433         # compute the readout
 434
 435         qr = torch.einsum("ntc,hdc->nhtd", x_q[:, t0:t1], self.w_qr)
 436
 437         ar = torch.einsum(
 438             "nhtd,nltd->nhlt",
 439             qr,
 440             self.rec_k[:, :, t0:t1],
 441         ) / math.sqrt(self.w_qr.size(1))
 442
 443         ar = ar.softmax(dim=2)  # nhlt
 444
 445         ar = F.dropout(ar, self.attention_dropout, self.training)
 446
 447         y = torch.einsum(
 448             "nhlt,nltd->nthd",
 449             ar,
 450             self.rec_v[:, :, t0:t1],
 451         ).flatten(2)
 452
 453         self.cache_y[:, t0:t1] = y @ self.w_o
 454
 455         return BracketedSequence(self.cache_y, t0, t1 - t0, bs.init_cache)
 456
 457
 458 ##############################
 459
 460
 461 # Returns a tensor with an additional index at rank win_dim, that move
 462 # along the same dimension as dim, on a domain {0...win_size-1}, and
 463 # dim is restricted on a domain reduced by win_size-1 values.
 464
 465
 466 def moving_window(x, dim, win_dim, win_size):
 467     size, stride = x.size(), x.stride()
 468     size = size[:dim] + (size[dim] - win_size + 1,) + size[dim + 1 :]
 469     size = size[:win_dim] + (win_size,) + size[win_dim:]
 470     stride = stride[:win_dim] + (stride[dim],) + stride[win_dim:]
 471
 472     return x.as_strided(size=size, stride=stride)
 473
 474
 475 ##############################
 476
 477
 478 class Caterpillar(nn.Module):
 479     def __init__(
 480         self,
 481         dim_model,
 482         dim_qk,
 483         dim_v,
 484         nb_heads,
 485         caterpillar_length,
 486         caterpillar_height,
 487         attention_dropout=0.0,
 488         len_max=1e5,
 489         logger=print,
 490         **kwargs,
 491     ):
 492         super().__init__()
 493
 494         warnings.warn("Caterpillar", RuntimeWarning)
 495
 496         def randw(*d, amplitude=None):
 497             if amplitude is None:
 498                 amplitude = 1 / math.sqrt(d[-1])
 499             return nn.Parameter(amplitude * torch.randn(*d))
 500
 501         self.caterpillar_length = caterpillar_length
 502         self.caterpillar_height = caterpillar_height
 503         self.attention_dropout = attention_dropout
 504
 505         ######################################################################
 506         # sup_args
 507
 508         x = kwargs.get("gate_dropout")
 509         if x is None:
 510             self.proba_gate_dropout = 0.0
 511         else:
 512             self.proba_gate_dropout = float(x)
 513
 514         logger(f"self.proba_gate_dropout {self.proba_gate_dropout}")
 515
 516         x = kwargs.get("default_bg")
 517         if x is None:
 518             default_bg = -math.log(caterpillar_height - 1)
 519         else:
 520             default_bg = float(x)
 521
 522         logger(f"default_bg {default_bg}")
 523
 524         ######################################################################
 525
 526         self.w_G = randw(nb_heads, caterpillar_height, dim_model)
 527         self.b_G = nn.Parameter(torch.full((nb_heads, caterpillar_height), default_bg))
 528
 529         self.w_K = randw(nb_heads, dim_qk, dim_model)
 530         self.w_V = randw(nb_heads, dim_v, dim_model)
 531         self.w_Q = randw(nb_heads, dim_qk, dim_model)
 532         self.w_O = randw(dim_v * nb_heads, dim_model)
 533
 534         self.init_K_rec = randw(
 535             caterpillar_height,
 536             caterpillar_length,
 537             dim_qk,
 538         )
 539         self.init_V_rec = randw(
 540             caterpillar_height,
 541             caterpillar_length,
 542             dim_v,
 543         )
 544
 545     def reset_inner_loss(self):
 546         self.acc_attention = 0
 547         self.acc_nb = 0
 548
 549     def get_inner_loss(self):
 550         # warnings.warn("l2 regularization", RuntimeWarning)
 551         # return (self.acc_attention / self.acc_nb).pow(2).sum()
 552         return torch.tensor([0], device=self.w_Q.device)
 553
 554     def forward(self, bs):
 555         # Dimensions to make the source a bit clearer, that's needed
 556
 557         X, t0, t1 = bs.slice(), bs.first, bs.first + bs.nb
 558
 559         N = bs.x.size(0)
 560         T = bs.x.size(1)
 561         H = self.w_V.size(0)
 562         DV = self.w_V.size(1)
 563         DK = self.w_K.size(1)
 564         DM = self.w_O.size(1)
 565         R = self.caterpillar_height
 566         L = self.caterpillar_length
 567
 568         assert (
 569             t0 >= L and (t1 - t0) % L == 0
 570         ), f"bs.first should be greater than caterpillar_length, and bs.nb should be a multiple of caterpillar_length"
 571
 572         # We cache values to deal efficiently with auto-regression
 573
 574         if bs.init_cache:
 575             self.rec_V = X.new_zeros(N, R, T, DV)
 576             self.rec_K = X.new_zeros(N, R, T, DK)
 577             # We start the recurrent sequences with optimizable
 578             # initial values. No idea if it helps.
 579             self.rec_V[:, :, t0 - L : t0, :] = self.init_V_rec[None, :, :, :]
 580             self.rec_K[:, :, t0 - L : t0, :] = self.init_K_rec[None, :, :, :]
 581
 582             self.cache_Y = X.new_zeros(N, T, DM)
 583
 584         V = torch.einsum("ntc,hdc->nhtd", X, self.w_V)
 585         K = torch.einsum("ntc,hdc->nhtd", X, self.w_K)
 586
 587         ######################################################################
 588         # Compute the recurrent state
 589
 590         # This is the Gating sequence that modulates the storing of
 591         # the new key and value in the R pairs of the current
 592         # stack. There are R independent gating values, which means
 593         # that the current K/V may be stored in multiple pairs of the
 594         # recurrent state, or not at all.
 595
 596         G = (
 597             torch.einsum("ntc,hrc->nhrt", X, self.w_G) + self.b_G[None, :, :, None]
 598         ).sigmoid()
 599
 600         # Clip the gating to avoid values greater than 1 when several
 601         # heads hit the same row
 602
 603         G = G / G.sum(1, keepdim=True).clamp(min=1)
 604
 605         ######################################################################
 606
 607         def recurrence(G, V, K):
 608             # We prepare the arguments for the parallel scan
 609
 610             A = 1 - G.sum(1)
 611
 612             gated_V = torch.einsum("nhrt,nhtd->nrtd", G, V)
 613             gated_K = torch.einsum("nhrt,nhtd->nrtd", G, K)
 614
 615             # We start from cached values, which matters in inference
 616
 617             init_rec_V = self.rec_V[:, :, t0 - L : t0]
 618             init_rec_K = self.rec_K[:, :, t0 - L : t0]
 619
 620             # Here there is a trick: Since the stack at position t is
 621             # computed by updating that at position t-L, the parallel
 622             # scan operates with a period of L. To do so we split the
 623             # sequence indexing in two axes, the second of size L, and
 624             # run the parallel scan using the first as the sequence index.
 625
 626             A = A.unflatten(2, (-1, L))
 627             gated_V = gated_V.unflatten(2, (-1, L))
 628             gated_K = gated_K.unflatten(2, (-1, L))
 629
 630             next_V = pscan_dim(A, gated_V, init_rec_V, dim=2)
 631             next_K = pscan_dim(A, gated_K, init_rec_K, dim=2)
 632
 633             next_V = next_V.flatten(2, 3)
 634             next_K = next_K.flatten(2, 3)
 635
 636             return next_V, next_K
 637
 638         #################################################################
 639
 640         next_V, next_K = recurrence(G, V, K)
 641
 642         if self.training and self.proba_gate_dropout > 0.0:
 643             # G is NxHxRxT where r is the caterpillar's row.
 644
 645             warnings.warn("gate dropout", RuntimeWarning)
 646
 647             # kill = (
 648             # torch.rand(G.size(), device=G.device) <= self.proba_gate_dropout
 649             # ).float()
 650
 651             kill = (
 652                 torch.rand(N, H, R, t1 - t0, device=G.device).sort(dim=3).indices == 0
 653             ).cumsum(dim=3)
 654             kill = kill * (
 655                 torch.rand(N, H, R, 1, device=G.device) <= self.proba_gate_dropout
 656             )
 657
 658             mask = 1 - kill
 659
 660             masked_next_V, masked_next_K = recurrence(G * mask, V, K)
 661
 662             next_V = next_V.detach() + (masked_next_V - masked_next_V.detach()) / (
 663                 1 - self.proba_gate_dropout
 664             )
 665             next_K = next_K.detach() + (masked_next_K - masked_next_K.detach()) / (
 666                 1 - self.proba_gate_dropout
 667             )
 668
 669         self.rec_V[:, :, t0:t1] = next_V
 670         self.rec_K[:, :, t0:t1] = next_K
 671
 672         ######################################################################
 673         # compute the readout
 674
 675         Q = torch.einsum("ntc,hdc->nhtd", X, self.w_Q)
 676
 677         # We build tensors NxHxTxFxL where N is the sample index, H
 678         # the head, T the time, F the row in the caterpillar, and L
 679         # the column in the caterpillar
 680
 681         windowed_V = moving_window(
 682             self.rec_V[:, :, t0 - L + 1 : t1], dim=2, win_dim=3, win_size=L
 683         )
 684
 685         windowed_K = moving_window(
 686             self.rec_K[:, :, t0 - L + 1 : t1], dim=2, win_dim=3, win_size=L
 687         )
 688
 689         # We have an attention score for each of the RxL values
 690
 691         ar = torch.einsum(
 692             "nhtd,nftld->nhtfl",
 693             Q,
 694             windowed_K,
 695         ) / math.sqrt(DK)
 696
 697         # softmax can operate only on one dimension, hence the
 698         # flattening
 699
 700         ar = ar.flatten(3).softmax(dim=3).view(ar.size())
 701
 702         ar = F.dropout(ar, self.attention_dropout, self.training)
 703
 704         # Compute the output for each head, flatten to concatenate
 705
 706         Y = torch.einsum(
 707             "nhtfl,nftld->nthd",
 708             ar,
 709             windowed_V,
 710         ).flatten(2)
 711
 712         # Compute the final output
 713
 714         self.cache_Y[:, t0:t1] = Y @ self.w_O
 715
 716         return BracketedSequence(self.cache_Y, t0, t1 - t0, bs.init_cache)
 717
 718
 719 ##############################
 720
 721
 722 class QKVAttention(nn.Module):
 723     def __init__(
 724         self,
 725         dim_model,
 726         dim_qk,
 727         dim_v,
 728         nb_heads=1,
 729         causal=False,
 730         attention_dropout=0.0,
 731         logger=print,
 732         **kwargs,
 733     ):
 734         super().__init__()
 735
 736         def randw(*d):
 737             return nn.Parameter(torch.randn(*d) / math.sqrt(d[-1]))
 738
 739         self.causal = causal
 740         self.attention_dropout = attention_dropout
 741         self.record_attention = False
 742
 743         self.w_q = randw(nb_heads, dim_qk, dim_model)
 744         self.w_k = randw(nb_heads, dim_qk, dim_model)
 745         self.w_v = randw(nb_heads, dim_v, dim_model)
 746         self.w_o = randw(dim_v * nb_heads, dim_model)
 747
 748     def forward(self, bs):
 749         x_q = bs.x
 750
 751         assert (
 752             self.causal or bs.complete()
 753         ), "Partial evaluation is only possible for causal models"
 754
 755         if bs.init_cache:
 756             self.cache_k = x_q.new_zeros(
 757                 x_q.size(0), self.w_k.size(0), x_q.size(1), self.w_k.size(1)
 758             )
 759             self.cache_v = x_q.new_zeros(
 760                 x_q.size(0), self.w_v.size(0), x_q.size(1), self.w_v.size(1)
 761             )
 762             self.cache_y = x_q.new_zeros(x_q.size(0), x_q.size(1), self.w_o.size(1))
 763
 764         q = torch.einsum("ntc,hdc->nhtd", x_q[:, bs.first : bs.first + bs.nb], self.w_q)
 765
 766         self.cache_k[:, :, bs.first : bs.first + bs.nb] = torch.einsum(
 767             "ntc,hdc->nhtd", x_q[:, bs.first : bs.first + bs.nb], self.w_k
 768         )
 769         self.cache_v[:, :, bs.first : bs.first + bs.nb] = torch.einsum(
 770             "ntc,hdc->nhtd", x_q[:, bs.first : bs.first + bs.nb], self.w_v
 771         )
 772
 773         a = torch.einsum(
 774             "nhtd,nhsd->nhts", q, self.cache_k[:, :, : bs.first + bs.nb]
 775         ) / math.sqrt(self.w_q.size(1))
 776
 777         if self.causal:
 778             if bs.init_cache:
 779                 self.cache_attzero = (
 780                     torch.arange(x_q.size(1), device=q.device)[None, None, :, None]
 781                     < torch.arange(x_q.size(1), device=q.device)[None, None, None, :]
 782                 )
 783             a = a.masked_fill(
 784                 self.cache_attzero[
 785                     :, :, bs.first : bs.first + bs.nb, : bs.first + bs.nb
 786                 ],
 787                 float("-inf"),
 788             )
 789
 790         a = a.softmax(dim=3)
 791
 792         if self.record_attention:
 793             self.a = a
 794
 795         a = F.dropout(a, self.attention_dropout, self.training)
 796
 797         y = torch.einsum(
 798             "nhts,nhsd->nthd", a, self.cache_v[:, :, : bs.first + bs.nb]
 799         ).flatten(2)
 800
 801         self.cache_y[:, bs.first : bs.first + bs.nb] = y @ self.w_o
 802
 803         return BracketedSequence(self.cache_y, bs.first, bs.nb, bs.init_cache)
 804
 805
 806 ##############################
 807
 808
 809 class MyGPT(nn.Module):
 810     def __init__(
 811         self,
 812         vocabulary_size,
 813         dim_model,
 814         dim_keys,
 815         dim_hidden,
 816         nb_heads,
 817         nb_blocks,
 818         nb_lines=None,
 819         caterpillar_height=None,
 820         causal=False,
 821         dropout=0.0,
 822         len_max=1e5,
 823         attention_layer="kvrec",
 824         logger=print,
 825         **kwargs,
 826     ):
 827         super().__init__()
 828
 829         assert attention_layer in {
 830             "mha",
 831             "dumbrec",
 832             "kvrec",
 833             "caterpillar",
 834         }, f"Unknown attention operator {attention_layer}."
 835
 836         if attention_layer == "caterpillar":
 837             assert nb_lines % caterpillar_height == 0
 838             self.caterpillar_length = nb_lines // caterpillar_height
 839             self.caterpillar_height = caterpillar_height
 840         else:
 841             self.caterpillar_length = -1
 842             self.caterpillar_height = -1
 843
 844         assert dim_model % nb_heads == 0
 845
 846         self.embedding = nn.Sequential(
 847             CacheWrapper(nn.Embedding(vocabulary_size, dim_model), nn.Dropout(dropout)),
 848             AddPositionalEncoding(len_max),
 849         )
 850
 851         trunk_blocks = []
 852
 853         def attlayer():
 854             if attention_layer == "mha":
 855                 return QKVAttention(
 856                     dim_model=dim_model,
 857                     dim_qk=dim_keys,
 858                     dim_v=dim_model // nb_heads,
 859                     nb_heads=nb_heads,
 860                     causal=causal,
 861                     attention_dropout=dropout,
 862                     logger=logger,
 863                     **kwargs,
 864                 )
 865             elif attention_layer == "dumbrec":
 866                 return DumbRec(
 867                     dim_model=dim_model,
 868                     dim_qk=dim_keys,
 869                     dim_v=dim_model // nb_heads,
 870                     nb_heads=nb_heads,
 871                     nb_lines=nb_lines,
 872                     attention_dropout=dropout,
 873                     logger=logger,
 874                     **kwargs,
 875                 )
 876             elif attention_layer == "kvrec":
 877                 return KVRec(
 878                     dim_model=dim_model,
 879                     dim_qk=dim_keys,
 880                     dim_v=dim_model // nb_heads,
 881                     nb_heads=nb_heads,
 882                     nb_lines=nb_lines,
 883                     attention_dropout=dropout,
 884                     logger=logger,
 885                     **kwargs,
 886                 )
 887             elif attention_layer == "caterpillar":
 888                 return Caterpillar(
 889                     dim_model=dim_model,
 890                     dim_qk=dim_keys,
 891                     dim_v=dim_model // nb_heads,
 892                     nb_heads=nb_heads,
 893                     caterpillar_length=self.caterpillar_length,
 894                     caterpillar_height=self.caterpillar_height,
 895                     attention_dropout=dropout,
 896                     logger=logger,
 897                     **kwargs,
 898                 )
 899             else:
 900                 raise ValueError(f"Unknown attention type {attention_layer}.")
 901
 902         for b in range(nb_blocks):
 903             trunk_blocks += [
 904                 WithResidual(
 905                     CacheWrapper(nn.LayerNorm((dim_model,))),
 906                     attlayer(),
 907                 ),
 908                 WithResidual(
 909                     CacheWrapper(
 910                         nn.LayerNorm((dim_model,)),
 911                         nn.Linear(in_features=dim_model, out_features=dim_hidden),
 912                         nn.ReLU(),
 913                         nn.Linear(in_features=dim_hidden, out_features=dim_model),
 914                         nn.Dropout(dropout),
 915                     ),
 916                 ),
 917             ]
 918
 919         self.trunk = nn.Sequential(*trunk_blocks)
 920
 921         self.readout = CacheWrapper(
 922             nn.Linear(in_features=dim_model, out_features=vocabulary_size)
 923         )
 924
 925         with torch.no_grad():
 926             for m in self.modules():
 927                 if isinstance(m, nn.Embedding):
 928                     m.weight.normal_(mean=0, std=2e-2)
 929                 elif isinstance(m, nn.LayerNorm):
 930                     m.bias.zero_()
 931                     m.weight.fill_(1.0)
 932
 933         self.reset_inner_loss()
 934
 935     def forward(self, bs):
 936         bs = BracketedSequence(F.pad(bs.x, (1, -1)), bs.first, bs.nb, bs.init_cache)
 937
 938         # To make the code simpler in the Caterpillar layer, we pad
 939         # here. It's unclear if/how much it hurts computationaly by
 940         # increasing the sequence length for the other layers
 941
 942         if self.caterpillar_length > 0:
 943             original_nb = bs.nb
 944             if bs.nb % self.caterpillar_length > 0:
 945                 bs.nb += self.caterpillar_length - bs.nb % self.caterpillar_length
 946
 947             bs = BracketedSequence(
 948                 F.pad(bs.x, (self.caterpillar_length, self.caterpillar_length)),
 949                 bs.first + self.caterpillar_length,
 950                 bs.nb,
 951                 bs.init_cache,
 952             )
 953
 954         bs = self.embedding(bs)
 955         bs = self.trunk(bs)
 956         bs = self.readout(bs)
 957
 958         if self.caterpillar_length > 0:
 959             bs = BracketedSequence(
 960                 F.pad(bs.x, (0, 0, -self.caterpillar_length, -self.caterpillar_length)),
 961                 bs.first - self.caterpillar_length,
 962                 original_nb,
 963                 bs.init_cache,
 964             )
 965
 966         return bs
 967
 968     # ar_mask is a tensor with 0s and 1s, of same shape as input, with
 969     # 1s where tokens should be generated. The others are kept
 970     # unchanged.
 971
 972     def masked_inplace_autoregression(
 973         self,
 974         input_src,
 975         ar_mask_src,
 976         forbidden_tokens=None,
 977         deterministic_synthesis=False,
 978     ):
 979         input = input_src.to(self.readout.f.weight.device)
 980         ar_mask = ar_mask_src.to(self.readout.f.weight.device)
 981         to_generate = (ar_mask.sum(0) > 0).nonzero()
 982         if to_generate.min() > 0:
 983             self(
 984                 BracketedSequence(input, 0, to_generate.min(), True)
 985             )  # Needed to initialize the model's cache
 986         for s in range(to_generate.min(), to_generate.max() + 1):
 987             output = self(BracketedSequence(input, s, 1, s == 0)).x
 988             logits = output[:, s]
 989             if forbidden_tokens is not None:
 990                 logits = logits.masked_fill(forbidden_tokens, float("-inf"))
 991             if deterministic_synthesis:
 992                 t_next = logits.argmax(1)
 993             else:
 994                 dist = torch.distributions.categorical.Categorical(logits=logits)
 995                 t_next = dist.sample()
 996             input[:, s] = ar_mask[:, s] * t_next + (1 - ar_mask[:, s]) * input[:, s]
 997
 998         input_src.copy_(input)
 999
1000     def reset_inner_loss(self):
1001         for m in self.modules():
1002             if m is not self and hasattr(m, "reset_inner_loss"):
1003                 m.reset_inner_loss()
1004
1005     def get_inner_loss(self):
1006         l = torch.tensor([0.0], device=self.readout.f.weight.device)
1007         for m in self.modules():
1008             if m is not self and hasattr(m, "get_inner_loss"):
1009                 l += m.get_inner_loss()
1010         return l
1011
1012     def record_attention(self, v=True):
1013         for m in self.modules():
1014             if isinstance(m, QKVAttention):
1015                 m.record_attention = v
1016
1017     def retrieve_attention(self):
1018         a = []
1019         for m in self.modules():
1020             if isinstance(m, QKVAttention):
1021                 a.append(m.a)
1022         return a
1023
1024
1025 ######################################################################
1026
1027 if __name__ == "__main__":
1028     print("Basic check.")
1029
1030     m = Caterpillar(
1031         dim_model=4,
1032         dim_qk=3,
1033         dim_v=7,
1034         nb_heads=1,
1035         caterpillar_length=7,
1036         caterpillar_height=3,
1037         attention_dropout=0.0,
1038     )
1039
1040     m.reset_inner_loss()
1041     x = torch.randn(1, 21 + 2 * 7, 4)
1042     y1 = m(BracketedSequence(x, first=7, nb=21, init_cache=True)).x[:, 7:28]
1043     y2 = m(BracketedSequence(x, first=7, nb=21, init_cache=True)).x[:, 7:28]
1044     y3a = m(BracketedSequence(x, first=7, nb=14, init_cache=True)).x[:, 7:21]
1045     y3b = m(BracketedSequence(x, first=21, nb=7, init_cache=False)).x[:, 21:28]
1046     print((y1 - y2).abs().max())
1047     print((y1 - torch.cat([y3a, y3b], dim=1)).abs().max())
1048     exit(0)
1049
1050     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
1051
1052     vocabulary_size = 128
1053     x = torch.randint(vocabulary_size, (6, 1024))
1054
1055     model = MyGPT(
1056         vocabulary_size=vocabulary_size,
1057         dim_model=512,
1058         dim_keys=64,
1059         dim_hidden=2048,
1060         nb_heads=8,
1061         nb_lines=128,
1062         nb_blocks=12,
1063         dropout=0.1,
1064         causal=True,
1065     )
1066
1067     x = x.to(device)
1068     model.to(device)
1069
1070     import time, sys
1071
1072     # import torchvision.models as models
1073     # from torch.profiler import profile, record_function, ProfilerActivity
1074
1075     # with profile(activities=[ProfilerActivity.CPU,  ProfilerActivity.CUDA], profile_memory=True, record_shapes=True) as prof:
1076     # with record_function("model_inference"):
1077
1078     model.eval()
1079     for i in range(3):
1080         start_time = time.perf_counter()
1081         for k in range(10):
1082             model(BracketedSequence(x))
1083         duration = time.perf_counter() - start_time
1084         print(duration)
1085         sys.stdout.flush()
1086
1087     # print(prof.key_averages().table(sort_by="cpu_time_total", row_limit=10))
1088     # print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=10))
1089
1090     # print("##############################################################")
1091     # y2 = torch.randn_like(y1)
1092     # for s in range(x.size(1)):
1093     # z = model(BracketedSequence(x, s, 1))
1094     # y2[:, s : s + 1] = z.slice()
1095
1096     # print(f"error={((y1 - y2).norm() / (y1.norm() + y2.norm())).item()}")
1097
1098 ######################################################################