-
Notifications
You must be signed in to change notification settings - Fork 1
/
cola.py
381 lines (341 loc) · 16.7 KB
/
cola.py
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
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
import math
import inspect
from dataclasses import dataclass, field
import numpy as np
from tqdm import tqdm
import torch
import torch.nn as nn
from torch.nn import functional as F
def new_gelu(x):
return 0.5 * x * (1.0 + torch.tanh(math.sqrt(2.0 / math.pi) * (x + 0.044715 * torch.pow(x, 3.0))))
class LayerNorm(nn.Module):
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 AttnMLP(nn.Module):
def __init__(self, config):
super().__init__()
self.lin_layers = nn.ModuleList([nn.Linear(config.n_embd, config.n_embd, bias=True) for _ in range(config.n_linear)])
self.dropout = nn.Dropout(config.dropout)
self.ln_layers = nn.ModuleList([LayerNorm(config.n_embd, bias=config.bias) for _ in range(config.n_linear)])
def forward(self, x):
for linear, ln in zip(self.lin_layers, self.ln_layers):
x = new_gelu(linear(ln(x)))
return x
class CausalSelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
assert config.n_embd % config.n_head == 0
# proj function
self.value_mlp = AttnMLP(config)
self.query_and_key_mlp = AttnMLP(config)
# Q, K, V transformation
self.query_transform = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
self.key_transform = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
self.value_transform = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
# output projection
self.c_proj = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
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
# support 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")
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()
unshare_x = self.value_mlp(x)
share_x = self.query_and_key_mlp(x)
q = self.query_transform(share_x)
k = self.key_transform(share_x)
v = self.value_transform(unshare_x)
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 (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
if self.flash:
y = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=self.dropout, is_causal=True)
else:
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)
# 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.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 = new_gelu(x)
x = self.c_proj(x)
x = self.dropout(x)
return x
class Block(nn.Module):
def __init__(self, config):
super().__init__()
self.ln_1 = LayerNorm(config.n_embd, bias=config.bias)
self.attn = CausalSelfAttention(config)
self.ln_2 = LayerNorm(config.n_embd, bias=config.bias)
self.mlp = MLP(config)
def forward(self, x):
x = x + self.attn(self.ln_1(x))
x = x + self.mlp(self.ln_2(x))
return x
@dataclass
class COLAConfig:
seed: int = 0
data: str = ''
datapath: str = ''
min_seq_len: int = 6
max_seq_len: int = 24
use_start_letter: bool = True
start_letter: int = 0
device: torch.device = torch.device('cuda:0')
block_size: int = 24
vocab_size: int = 20000
token_size: int = 20001
n_linear: int = 1
n_layer: int = 2
n_head: int = 2
n_embd: int = 96
dropout: float = 0.0
bias: bool = True
meta_lr: float = 1e-3
update_lr: float = 0.01
meta_epochs: int= 5
city_epochs: int = 1
test_epochs: int = 50
domain_specific_params: list = field(default_factory=lambda:['value_mlp', 'value_transform', 'wte'])
def to_dict(self):
return {k: v for k, v in self.__dict__.items()}
class COLA(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
if self.config.use_start_letter:
self.transformer = nn.ModuleDict(dict(
wte = nn.Embedding(config.token_size, config.n_embd, padding_idx=config.token_size-1),
wpe = nn.Embedding(config.block_size, config.n_embd),
drop = nn.Dropout(config.dropout),
h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
ln_f = LayerNorm(config.n_embd, bias=config.bias),
))
self.lm_head = nn.Linear(config.n_embd, config.token_size, bias=False)
else:
self.transformer = nn.ModuleDict(dict(
wte = nn.Embedding(config.vocab_size, config.n_embd),
wpe = nn.Embedding(config.block_size, config.n_embd),
drop = nn.Dropout(config.dropout),
h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
ln_f = LayerNorm(config.n_embd, bias=config.bias),
))
self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
# !!! ATTENTION !!!
# duplicate parameters will not be included in 'named_parameters', i.e., the parameters of lm_head are private as same as wte and not involved in meta updating.
# if changing the order of initialization, SPEC_DICT['sharemlp'] should add 'lm_head' to ensure the wte parameters (copy from lm_head) are not shared by all cities.
self.transformer.wte.weight = self.lm_head.weight
self.apply(self._init_weights)
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))
print("number of parameters: %.2fM" % (self.get_num_params()/1e6,))
def get_num_params(self, non_embedding=True):
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 copy_invariant_params(self, city_model):
for (m_name, m_param), (c_name, c_param) in zip(self.named_parameters(), city_model.named_parameters()):
# !!! ATTENTION !!!
contains_specific = any(sub_str in m_name for sub_str in self.config.domain_specific_params)
if not contains_specific:
assert m_name == c_name
c_param.data = m_param.data.clone()
assert torch.allclose(c_param.data, m_param.data)
def get_acc_topk(self, preds, targets):
acc_K = [1, 5, 10, 20]
result = {}
totalMRR = []
for K in acc_K:
result[K] = 0
seq_len_l = []
for i in range(len(preds)):
max_len = self.config.max_seq_len if self.config.use_start_letter else self.config.max_seq_len - 1
seq_len = max_len - len(torch.where(targets[i]==-1)[0])
seq_len_l.append(seq_len)
for j in range(seq_len):
pred, target = preds[i][j], targets[i][j].item()
sortedPred = torch.topk(pred, len(pred))[1].tolist()
truthIndex = sortedPred.index(target) + 1
avgPrec = 1 / truthIndex
totalMRR.append(avgPrec)
sorted_indexs = {}
for K in acc_K:
sorted_indexs[K] = sortedPred[:K]
if target in sorted_indexs[K]:
result[K] += 1
result['num_of_test'] = sum(seq_len_l)
result['mrr'] = np.sum(totalMRR)
result['mrr_num'] = len(totalMRR)
return result
def forward(self, idx, targets=None, freqs=None, use_acc = False):
device = idx.device
b, t = idx.size()
assert t <= self.config.block_size, f"Cannot forward sequence of length {t}, block size is only {self.config.block_size}"
pos = torch.arange(0, t, dtype=torch.long, device=device).unsqueeze(0)
tok_emb = self.transformer.wte(idx)
pos_emb = self.transformer.wpe(pos)
x = self.transformer.drop(tok_emb + pos_emb)
for block in self.transformer.h:
x = block(x)
x = self.transformer.ln_f(x)
if targets is not None:
logits = self.lm_head(x)
if self.config.use_start_letter:
logits[:, :, -1] = float('-inf')
loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.reshape(-1), ignore_index=-1)
else:
logits = self.lm_head(x[:, [-1], :])
if self.config.use_start_letter:
logits[:, :, -1] = float('-inf')
loss = None
if use_acc:
accs = self.get_acc_topk(logits, targets)
return logits, loss, accs
else:
return logits, loss
def crop_block_size(self, block_size):
assert block_size <= self.config.block_size
self.config.block_size = block_size
self.transformer.wpe.weight = nn.Parameter(self.transformer.wpe.weight[:block_size])
for block in self.transformer.h:
block.attn.bias = block.attn.bias[:,:,:block_size,:block_size]
def configure_optimizers(self, weight_decay, learning_rate, betas, device_type):
decay = set()
no_decay = set()
whitelist_weight_modules = (torch.nn.Linear, )
blacklist_weight_modules = (torch.nn.LayerNorm, LayerNorm, torch.nn.Embedding)
for mn, m in self.named_modules():
for pn, p in m.named_parameters():
fpn = '%s.%s' % (mn, pn) if mn else pn
if pn.endswith('bias'):
no_decay.add(fpn)
elif pn.endswith('weight') and isinstance(m, whitelist_weight_modules):
decay.add(fpn)
elif pn.endswith('weight') and isinstance(m, blacklist_weight_modules):
no_decay.add(fpn)
decay.remove('lm_head.weight')
param_dict = {pn: p for pn, p in self.named_parameters()}
inter_params = decay & no_decay
union_params = decay | no_decay
assert len(inter_params) == 0, "parameters %s made it into both decay/no_decay sets!" % (str(inter_params), )
assert len(param_dict.keys() - union_params) == 0, "parameters %s were not separated into either decay/no_decay set!" \
% (str(param_dict.keys() - union_params), )
optim_groups = [
{"params": [param_dict[pn] for pn in sorted(list(decay))], "weight_decay": weight_decay},
{"params": [param_dict[pn] for pn in sorted(list(no_decay))], "weight_decay": 0.0},
]
use_fused = (device_type == 'cuda') and ('fused' in inspect.signature(torch.optim.AdamW).parameters)
print(f"using fused AdamW: {use_fused}")
extra_args = dict(fused=True) if use_fused else dict()
optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=betas, **extra_args)
return optimizer
def estimate_mfu(self, fwdbwd_per_iter, dt):
N = self.get_num_params()
cfg = self.config
L, H, Q, T = cfg.n_layer, cfg.n_head, cfg.n_embd//cfg.n_head, cfg.block_size
flops_per_token = 6*N + 12*L*H*Q*T
flops_per_fwdbwd = flops_per_token * T
flops_per_iter = flops_per_fwdbwd * fwdbwd_per_iter
flops_achieved = flops_per_iter * (1.0/dt)
flops_promised = 312e12
mfu = flops_achieved / flops_promised
return mfu
@torch.no_grad()
def generate(self, args, len_list, num_samples, temperature=1.0, top_k=None):
"""
Take a conditioning sequence of indices idx (LongTensor of shape (b,t)) and complete
the sequence max_new_tokens times, feeding the predictions back into the model each time.
Most likely you'll want to make sure to be in model.eval() mode of operation for this.
"""
args.freqs = args.freqs / args.freqs.sum()
adjustments = np.log(args.freqs ** args.balance_coef + 1e-12)
adjustments = torch.from_numpy(adjustments)
adjustments = adjustments.to(args.device)
len_vals, len_cnts = np.unique(len_list, return_counts=True)
len_cnts = len_cnts / np.sum(len_cnts)
samples_len = [np.random.choice(len_vals, 1, p = len_cnts)[0] for i in range(num_samples)]
if self.config.use_start_letter:
idx = torch.LongTensor([self.config.start_letter]*num_samples).reshape(-1, 1).to(args.device)
gen_seq_len = self.config.max_seq_len
get_seq_offset = 1
else:
start_dist=torch.tensor(np.load(f'{self.config.datapath}/{self.config.data}/start.npy')).float()
idx = torch.LongTensor([torch.multinomial(start_dist, 1) for _ in range(num_samples)]).reshape(-1, 1).to(args.device)
gen_seq_len = self.config.max_seq_len - 1
get_seq_offset = 0
for t in range(gen_seq_len):
idx_cond = idx if idx.size(1) <= self.config.block_size else idx[:, -self.config.block_size:]
logits, _ = self(idx_cond)
# adjust logits when generating
logits = logits - adjustments
logits = logits[:, -1, :] / temperature
if top_k is not None:
v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
logits[logits < v[:, [-1]]] = -float('Inf')
probs = F.softmax(logits, dim=-1)
idx_next = torch.multinomial(probs, num_samples=1)
idx = torch.cat((idx, idx_next), dim=1)
pred = []
for i in range(len(idx)):
seq = idx[i][get_seq_offset:samples_len[i]+get_seq_offset]
pred.append(list(seq.cpu().numpy()))
return pred
@torch.no_grad()
def generate_for_seir(self, args, num_samples, temperature=1.0, top_k=None):
args.freqs = args.freqs / args.freqs.sum()
adjustments = np.log(args.freqs ** args.balance_coef + 1e-12)
adjustments = torch.from_numpy(adjustments)
adjustments = adjustments.to(args.device)
start_dist=torch.tensor(np.load(f'{self.config.datapath}/{self.config.data}/start.npy')).float()
idx = torch.LongTensor([torch.multinomial(start_dist, 1) for _ in range(num_samples)]).reshape(-1, 1).to(args.device)
gen_seq_len = 24*7-1
for t in tqdm(range(gen_seq_len)):
idx_cond = idx if idx.size(1) <= self.config.block_size else idx[:, -self.config.block_size:]
logits, _ = self(idx_cond)
# adjust logits when generating
logits = logits - adjustments
logits = logits[:, -1, :] / temperature
if top_k is not None:
v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
logits[logits < v[:, [-1]]] = -float('Inf')
probs = F.softmax(logits, dim=-1)
idx_next = torch.multinomial(probs, num_samples=1)
idx = torch.cat((idx, idx_next), dim=1)
pred = []
for i in range(len(idx)):
seq = idx[i][:]
pred.append(list(seq.cpu().numpy()))
return pred