-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_graph.py
134 lines (107 loc) · 4 KB
/
train_graph.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
import os, sys
import warnings
warnings.filterwarnings('ignore', category=FutureWarning)
import numpy as np
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torch.nn import DataParallel
from torch.optim import Adam
from torch.optim.lr_scheduler import StepLR
from torch.utils.tensorboard import SummaryWriter
from torchvision.transforms import Compose
from tqdm import tqdm
import argparse
from rplan import RrplanGraph, Flip, Rot90
from gpt2 import GraphGPTModel
from transformers.configuration_gpt2 import GPT2Config
if __name__ == '__main__':
dset = RrplanGraph(root_dir='/mnt/iscratch/datasets/rplan_ddg_var',
split='train',
seq_len=120,
edg_len=100,
vocab_size=65)
dloader = DataLoader(dset, batch_size=64, num_workers=10, shuffle=True)
val_set = RrplanGraph(root_dir='/mnt/iscratch/datasets/rplan_ddg_var',
split='val',
seq_len=120,
edg_len=100,
vocab_size=65)
val_loader = DataLoader(val_set, batch_size=64, num_workers=10, shuffle=True)
enc = GPT2Config(
vocab_size=65,
n_positions=120,
n_ctx=120,
n_embd=264,
n_layer=12,
n_head=12,
is_causal=False,
is_encoder=True,
id_embed=True
)
dec = GPT2Config(
vocab_size=65,
n_positions=100,
n_ctx=100,
n_embd=264,
n_layer=12,
n_head=12,
is_causal=True,
is_encoder=False
)
model = GraphGPTModel(enc, dec)
model = DataParallel(model.cuda())
optimizer = Adam(model.parameters(), lr=1e-4, eps=1e-6)
lr_scheduler = StepLR(optimizer, step_size=15, gamma=0.1)
writer = SummaryWriter(comment='id_but_otherwise_same_as_12_model')
global_steps = 1
val_steps = 1
for epochs in range(40):
model.train()
for steps, data in tqdm(enumerate(dloader)):
global_steps += 1
optimizer.zero_grad()
vert_seq = data['vert_seq'].cuda()
edg_seq = data['edg_seq'].cuda()
attn_mask = data['attn_mask'].cuda()
pos_id = data['pos_id'].cuda()
vert_attn_mask = data['vert_attn_mask'].cuda()
# print(vert_seq.shape)
loss = model( node=vert_seq,
edg=edg_seq,
attention_mask=attn_mask,
labels=edg_seq,
vert_attn_mask=vert_attn_mask)
# print(len(loss))
# for v in loss:
# if isinstance(v, torch.Tensor):
# print(v.shape)
# else:
# for vv in v:
# print('\t', vv.shape)
# print(loss[1])
loss[0].mean().backward()
optimizer.step()
# if steps % 100 == 0:
writer.add_scalar('loss/train', loss[0].mean(), global_step=global_steps)
torch.save(model.state_dict(), f'id_embed_12_modelv_eps_m6_mlp_lr_m4_{epochs}.pth')
# torch.save(model.state_dict(), f'face_modelv_eps_m6_mlp_lr_m4_{epochs}.pth')
lr_scheduler.step()
model.eval()
val_step_size = (global_steps - val_steps) // len(val_loader)
with torch.no_grad():
for steps, data in tqdm(enumerate(val_loader)):
vert_seq = data['vert_seq'].cuda()
edg_seq = data['edg_seq'].cuda()
attn_mask = data['attn_mask'].cuda()
pos_id = data['pos_id'].cuda()
vert_attn_mask = data['vert_attn_mask'].cuda()
loss = model( node=vert_seq,
edg=edg_seq,
attention_mask=attn_mask,
labels=edg_seq,
vert_attn_mask=vert_attn_mask)
writer.add_scalar('loss/val', loss[0].mean(), global_step=val_steps)
val_steps += val_step_size
writer.close()
'triples_hw3_best.pth'