-
Notifications
You must be signed in to change notification settings - Fork 0
/
tree.py
268 lines (240 loc) · 10.5 KB
/
tree.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
import os
import sys
import time
if (slurm_submit_dir := os.environ.get('SLURM_SUBMIT_DIR', default=None)) is not None:
sys.path.append(os.environ['SLURM_SUBMIT_DIR'])
import argparse
import torch
from eval.tasks.tree import TreeCopy, TreeReorder, C3, TreeApply
from eval.tasks.tree.batching import make_collator
from eval.models.tree import TreeUnitary, ShivQuirk, Model
from unitaryPE.nn.schedule import make_schedule
from torch.distributions import Normal
from torch.utils.data import DataLoader
from torch.optim import AdamW
from torch.optim.lr_scheduler import LambdaLR
from typing import Literal
def run(
model: Model,
vocab_size: int,
tree_depth_mu: int,
tree_depth_var: int,
task: Literal['copy', 'reorder', 'c3', 'apply'],
num_epochs: int,
num_layers: tuple[int, int],
num_heads: int,
dim: int,
regression: Literal['breadth', 'depth'],
store_path: str | None,
seed: int = 42):
start_time = time.time()
match task:
case 'copy':
task = TreeCopy(vocab_size=vocab_size, x_projection='breadth', y_projection=regression, sos_token_id=0, eos_token_id=-1)
case 'reorder':
task = TreeReorder(vocab_size=vocab_size, x_projection='breadth', y_projection=regression, sos_token_id=0, eos_token_id=-1)
case 'c3':
task = C3(x_projection='breadth', y_projection=regression, sos_token_id=0, eos_token_id=-1)
case 'apply':
task = TreeApply(x_projection='breadth', y_projection=regression, vocab_size=vocab_size, sos_token_id=0, eos_token_id=-1)
case _:
raise ValueError
train_depth_dist = Normal(tree_depth_mu, tree_depth_var)
test_depth_dist = Normal(tree_depth_mu, tree_depth_var)
train_set, dev_set, _ = task.make_sets(
distributions=(train_depth_dist, train_depth_dist, test_depth_dist),
num_samples=(6000, 2000, 2000),
seed=42) # keep this fixed for data consistency
print(sum(t.x.numel() for t in train_set)/len(train_set))
print(sum(t.y.numel() for t in train_set) / len(train_set))
train_dl = DataLoader([sample.process() for sample in train_set], # noqa
batch_size=64, collate_fn=make_collator('cuda'), shuffle=True)
dev_dl = DataLoader([sample.process() for sample in dev_set], # noqa
batch_size=32, collate_fn=make_collator('cuda'), shuffle=False)
torch.manual_seed(seed)
match model:
case Model.Unitary:
model = TreeUnitary(
vocab_size=vocab_size + 2,
dim=dim,
num_heads=num_heads,
num_layers=num_layers,
branching_factor=2).to('cuda')
case Model.ShivQuirk:
model = ShivQuirk(
vocab_size=vocab_size + 2,
dim=dim,
num_heads=num_heads,
num_layers=num_layers,
branching_factor=2,
max_depth=tree_depth_mu + tree_depth_var).to('cuda')
case _:
raise ValueError
steps_per_epoch = len(train_dl)
optim = AdamW(model.parameters(), lr=1)
scheduler = LambdaLR(
optimizer=optim,
lr_lambda=make_schedule(
warmup_steps=steps_per_epoch * 5,
warmdown_steps=steps_per_epoch * (num_epochs - 5),
total_steps=steps_per_epoch * num_epochs,
min_lr=1e-9,
max_lr=5e-4,
init_lr=1e-7))
best_epoch, best_dev_loss = None, 1e10
for epoch in range(num_epochs):
epoch_loss = 0.
model.train()
print(f'{epoch}')
for ((input_ids, input_pos, input_mask), (output_ids, output_pos, output_mask), causal_mask) in train_dl:
loss = model.get_loss(
input_ids=input_ids,
input_pos=input_pos,
input_mask=input_mask,
output_ids=output_ids,
output_pos=output_pos,
causal_mask=causal_mask,
label_smoothing=0.)
epoch_loss += loss.item()
loss.backward()
optim.step()
scheduler.step()
optim.zero_grad()
print(f'Train loss {epoch_loss}')
model.eval()
epoch_loss = 0.
if (epoch > 0 and epoch % 5 == 0) or epoch > num_epochs // 2:
with torch.no_grad():
for ((input_ids, input_pos, input_mask), (output_ids, output_pos, _), causal_mask) in dev_dl:
loss = model.get_loss(
input_ids=input_ids,
input_pos=input_pos,
input_mask=input_mask,
output_ids=output_ids,
output_pos=output_pos,
causal_mask=causal_mask,
label_smoothing=0.)
epoch_loss += loss.item()
print(f'Dev loss {epoch_loss}')
if epoch_loss < best_dev_loss and store_path is not None:
best_epoch, best_dev_loss = epoch, epoch_loss
torch.save(model.state_dict(), store_path)
duration = time.time() - start_time
print(f'Training took {duration} seconds. Best epoch was {best_epoch}')
sys.stdout.flush()
def evaluate(
model: Model,
vocab_size: int,
tree_depth_mu: int,
tree_depth_var: int,
task: Literal['copy', 'reorder', 'c3', 'apply'],
num_layers: tuple[int, int],
num_heads: int,
dim: int,
regression: Literal['breadth', 'depth'],
store_path: str | None,
seed: int = 42):
match task:
case 'copy':
task = TreeCopy(vocab_size=vocab_size, x_projection='breadth', y_projection=regression, sos_token_id=0, eos_token_id=-1)
case 'reorder':
task = TreeReorder(vocab_size=vocab_size, x_projection='breadth', y_projection=regression, sos_token_id=0, eos_token_id=-1)
case 'c3':
task = C3(x_projection='breadth', y_projection=regression, sos_token_id=0, eos_token_id=-1)
case 'apply':
task = TreeApply(x_projection='breadth', y_projection=regression, vocab_size=vocab_size, sos_token_id=0, eos_token_id=-1)
case _:
raise ValueError
train_depth_dist = Normal(tree_depth_mu, tree_depth_var)
test_depth_dist = Normal(tree_depth_mu, tree_depth_var)
_, _, test_set = task.make_sets(
distributions=(train_depth_dist, train_depth_dist, test_depth_dist),
num_samples=(6000, 2000, 2000),
seed=42) # keep this fixed for data consistency
test_dl = DataLoader([sample.process() for sample in test_set], # noqa
batch_size=32, collate_fn=make_collator('cuda'), shuffle=False)
torch.manual_seed(seed)
match model:
case Model.Unitary:
model = TreeUnitary(
vocab_size=vocab_size + 2,
dim=dim,
num_heads=num_heads,
num_layers=num_layers,
branching_factor=2).to('cuda')
case Model.ShivQuirk:
model = ShivQuirk(
vocab_size=vocab_size + 2,
dim=dim,
num_heads=num_heads,
num_layers=num_layers,
branching_factor=2,
max_depth=tree_depth_mu + tree_depth_var).to('cuda')
case _:
raise ValueError
model.load_state_dict(torch.load(store_path, map_location='cuda'), strict=True)
model.eval()
loss = torch.tensor([], device='cuda', dtype=torch.float)
with torch.no_grad():
for ((input_ids, input_pos, input_mask), (output_ids, output_pos, _), causal_mask) in test_dl:
pad_mask = output_ids[:, :-1].flatten().ne(-1)
xe = model.get_loss(
input_ids=input_ids,
input_pos=input_pos,
input_mask=input_mask,
output_ids=output_ids,
output_pos=output_pos,
causal_mask=causal_mask,
reduction='none',
label_smoothing=0.
)[pad_mask]
loss = torch.cat((loss, xe), dim=-1)
ppl = torch.exp(torch.mean(loss)).item()
print(f'{ppl=}')
sys.stdout.flush()
def parse_args():
parser = argparse.ArgumentParser(description='Run a single training iteration')
parser.add_argument('--eval', action='store_true')
parser.add_argument('--model', type=str, required=True, choices=['Unitary', 'ShivQuirk'], help='Type of model to use')
parser.add_argument('--regression', type=str, required=True, choices=['breadth', 'depth'], help='Decoding order')
parser.add_argument('--vocab_size', type=int, default=20, help='Size of vocabulary')
parser.add_argument('--tree_depth_mu', type=int, default=7, help='Mean tree depth')
parser.add_argument('--tree_depth_var', type=int, default=1, help='Tree depth variance')
parser.add_argument('--task', type=str, required=True, choices=['copy', 'reorder', 'c3', 'apply'], help='Which task to train on')
parser.add_argument('--num_epochs', type=int, default=400, help='Number of training epochs')
parser.add_argument('--num_layers', type=int, nargs=2, default=(2, 2), help='Number of layers for the model')
parser.add_argument('--dim', type=int, default=512, help='Dimension of the model')
parser.add_argument('--num_heads', type=int, default=8, help='Number of attention heads')
parser.add_argument('--store_path', type=str, default=None, help='If/where to store the trained model')
parser.add_argument('--seed', type=int, default=42, help='The id of the current repetition')
return parser.parse_args()
if __name__ == '__main__':
args = parse_args()
print(args)
if args.eval:
evaluate(
model=Model[args.model],
task=args.task,
num_heads=args.num_heads,
vocab_size=args.vocab_size,
tree_depth_mu=args.tree_depth_mu,
tree_depth_var=args.tree_depth_var,
dim=args.dim,
num_layers=args.num_layers,
store_path=args.store_path,
regression=args.regression,
seed=args.seed)
else:
run(
model=Model[args.model],
task=args.task,
num_heads=args.num_heads,
num_epochs=args.num_epochs,
vocab_size=args.vocab_size,
tree_depth_mu=args.tree_depth_mu,
tree_depth_var=args.tree_depth_var,
dim=args.dim,
num_layers=args.num_layers,
store_path=args.store_path,
regression=args.regression,
seed=args.seed)