Comments (7)
peterjc123
commented on May 2, 2024
@liamsun2019 Hi, thanks for trying out our project. With concern to this issue, I cannot say where the problem is without further information. I think there are several things you could do to help diagnose this problem.
- Check for the device of the parameters at critical code locations
for n, p in qat_model.named_parameters():
assert p.device == device, f"{n} is not on the desired device, but {p.device}"- Try evaluating once with the
dummy_inputgiven before passing intotrain
qat_model(dummy_input.to(device=device))BTW, it would be helpful if you could share your code (e.g. train_one_epoch , validate and the other parts of your main script) and the generated model description script.
from tinyneuralnetwork.
liamsun2019
commented on May 2, 2024
Thanks for your tips. I tried the 2 things you mentioned and both ran normally:
- The device of all the named parameters are cuda:0 as expected
- The evaluation looks normal
I attach the main.py and the generated model description script for your reference. As for validate/train_one_epoch, I will prepare them later when needed.
main.py:
def main_quant(opt):
torch.manual_seed(opt.seed)
torch.backends.cudnn.benchmark = not opt.not_cuda_benchmark and not opt.test
Dataset = get_dataset(opt.dataset, opt.task)
opt = opts().update_dataset_info_and_set_heads(opt, Dataset)
print(opt)
logger = Logger(opt)
os.environ['CUDA_VISIBLE_DEVICES'] = opt.gpus_str
opt.device = torch.device('cuda' if opt.gpus[0] >= 0 else 'cpu')
print('Creating model...')
if not opt.froze_backbone:
model = create_model(opt.arch, opt.heads,
opt.head_conv, opt.froze_backbone, opt.ft_size, norm_layer=nn.BatchNorm2d)
else:
model = create_model(opt.arch, opt.heads,
opt.head_conv, opt.froze_backbone, opt.ft_size)
device = get_device()
optimizer = torch.optim.Adam(model.parameters(), opt.lr)
start_epoch = 0
if opt.load_model != '':
model, optimizer, start_epoch = load_model(
model, opt.load_model, optimizer, opt.resume, opt.lr, opt.lr_step)
Trainer = train_factory[opt.task]
trainer = Trainer(opt, model, optimizer)
trainer.set_device(opt.gpus, opt.chunk_sizes, opt.device)
dummy_input = torch.rand((1, 3, opt.input_h, opt.input_w))
context = DLContext()
context.device = device
context.val_loader = torch.utils.data.DataLoader(
Dataset(opt, 'val'),
batch_size=1,
shuffle=False,
num_workers=1,
pin_memory=True
)
context.train_loader = torch.utils.data.DataLoader(
Dataset(opt, 'train'),
batch_size=opt.batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True,
drop_last=True
)
print("Start preparing the model for quantization")
quantizer = QATQuantizer(model, dummy_input, work_dir='out')
qat_model = quantizer.quantize()
print("Start quantization-aware training ===>", device)
qat_model.to(device=device)
context = DLContext()
context.device = device
context.val_loader = torch.utils.data.DataLoader(
Dataset(opt, 'val'),
batch_size=1,
shuffle=False,
num_workers=1,
pin_memory=True
)
context.train_loader = torch.utils.data.DataLoader(
Dataset(opt, 'train'),
batch_size=opt.batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True,
drop_last=True
)
context.max_epoch = 50
for n, p in qat_model.named_parameters():
assert p.device == device, f"{n} is not on the desired device, but {p.device}"
qat_model(dummy_input.to(device=device))
# Quantization-aware training
train(qat_model, context, trainer.train_one_epoch, trainer.validate, qat=True)
print("Start converting the model to TFLite")
logger.close()
from tinyneuralnetwork.
liamsun2019
commented on May 2, 2024
The generated model description script:
import torch
import torch.nn
import torch.functional
import torch.nn.functional
import torch.quantization
import torch.nn.quantized
class MoveNet_qat(torch.nn.Module):
def init(self):
super().init()
self.fake_quant_0 = torch.quantization.QuantStub()
self.backbone_body_0_0 = torch.nn.ZeroPad2d(padding=(0, 1, 0, 1))
self.backbone_body_0_1 = torch.nn.Conv2d(3, 56, kernel_size=(3, 3), stride=(2, 2))
self.backbone_body_0_2 = torch.nn.ReLU6(inplace=True)
self.backbone_body_1_conv_0_0 = torch.nn.Conv2d(56, 56, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=56)
self.backbone_body_1_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_1_conv_1 = torch.nn.Conv2d(56, 32, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_2_conv_0_0 = torch.nn.Conv2d(32, 192, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_2_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_2_conv_1_0 = torch.nn.ZeroPad2d(padding=(0, 1, 0, 1))
self.backbone_body_2_conv_1_1 = torch.nn.Conv2d(192, 192, kernel_size=(3, 3), stride=(2, 2), groups=192)
self.backbone_body_2_conv_1_2 = torch.nn.ReLU6(inplace=True)
self.backbone_body_2_conv_2 = torch.nn.Conv2d(192, 40, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_3_conv_0_0 = torch.nn.Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_3_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_3_conv_1_0 = torch.nn.Conv2d(240, 240, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=240)
self.backbone_body_3_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_3_conv_2 = torch.nn.Conv2d(240, 40, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_4_conv_0_0 = torch.nn.Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_4_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_4_conv_1_0 = torch.nn.ZeroPad2d(padding=(0, 1, 0, 1))
self.backbone_body_4_conv_1_1 = torch.nn.Conv2d(240, 240, kernel_size=(3, 3), stride=(2, 2), groups=240)
self.backbone_body_4_conv_1_2 = torch.nn.ReLU6(inplace=True)
self.backbone_body_4_conv_2 = torch.nn.Conv2d(240, 56, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_5_conv_0_0 = torch.nn.Conv2d(56, 336, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_5_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_5_conv_1_0 = torch.nn.Conv2d(336, 336, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=336)
self.backbone_body_5_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_5_conv_2 = torch.nn.Conv2d(336, 56, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_6_conv_0_0 = torch.nn.Conv2d(56, 336, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_6_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_6_conv_1_0 = torch.nn.Conv2d(336, 336, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=336)
self.backbone_body_6_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_6_conv_2 = torch.nn.Conv2d(336, 56, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_7_conv_0_0 = torch.nn.Conv2d(56, 336, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_7_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_7_conv_1_0 = torch.nn.ZeroPad2d(padding=(0, 1, 0, 1))
self.backbone_body_7_conv_1_1 = torch.nn.Conv2d(336, 336, kernel_size=(3, 3), stride=(2, 2), groups=336)
self.backbone_body_7_conv_1_2 = torch.nn.ReLU6(inplace=True)
self.backbone_body_7_conv_2 = torch.nn.Conv2d(336, 112, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_8_conv_0_0 = torch.nn.Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_8_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_8_conv_1_0 = torch.nn.Conv2d(672, 672, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=672)
self.backbone_body_8_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_8_conv_2 = torch.nn.Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_9_conv_0_0 = torch.nn.Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_9_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_9_conv_1_0 = torch.nn.Conv2d(672, 672, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=672)
self.backbone_body_9_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_9_conv_2 = torch.nn.Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_10_conv_0_0 = torch.nn.Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_10_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_10_conv_1_0 = torch.nn.Conv2d(672, 672, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=672)
self.backbone_body_10_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_10_conv_2 = torch.nn.Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_11_conv_0_0 = torch.nn.Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_11_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_11_conv_1_0 = torch.nn.Conv2d(672, 672, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=672)
self.backbone_body_11_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_11_conv_2 = torch.nn.Conv2d(672, 168, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_12_conv_0_0 = torch.nn.Conv2d(168, 1008, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_12_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_12_conv_1_0 = torch.nn.Conv2d(1008, 1008, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1008)
self.backbone_body_12_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_12_conv_2 = torch.nn.Conv2d(1008, 168, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_13_conv_0_0 = torch.nn.Conv2d(168, 1008, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_13_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_13_conv_1_0 = torch.nn.Conv2d(1008, 1008, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1008)
self.backbone_body_13_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_13_conv_2 = torch.nn.Conv2d(1008, 168, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_14_conv_0_0 = torch.nn.Conv2d(168, 1008, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_14_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_14_conv_1_0 = torch.nn.ZeroPad2d(padding=(0, 1, 0, 1))
self.backbone_body_14_conv_1_1 = torch.nn.Conv2d(1008, 1008, kernel_size=(3, 3), stride=(2, 2), groups=1008)
self.backbone_body_14_conv_1_2 = torch.nn.ReLU6(inplace=True)
self.backbone_body_14_conv_2 = torch.nn.Conv2d(1008, 280, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_15_conv_0_0 = torch.nn.Conv2d(280, 1680, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_15_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_15_conv_1_0 = torch.nn.Conv2d(1680, 1680, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1680)
self.backbone_body_15_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_15_conv_2 = torch.nn.Conv2d(1680, 280, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_16_conv_0_0 = torch.nn.Conv2d(280, 1680, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_16_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_16_conv_1_0 = torch.nn.Conv2d(1680, 1680, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1680)
self.backbone_body_16_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_16_conv_2 = torch.nn.Conv2d(1680, 280, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_17_conv_0_0 = torch.nn.Conv2d(280, 1680, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_17_conv_0_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_17_conv_1_0 = torch.nn.Conv2d(1680, 1680, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1680)
self.backbone_body_17_conv_1_1 = torch.nn.ReLU6(inplace=True)
self.backbone_body_17_conv_2 = torch.nn.Conv2d(1680, 560, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_18_0 = torch.nn.Conv2d(560, 1280, kernel_size=(1, 1), stride=(1, 1))
self.backbone_body_18_1 = torch.nn.ReLU6(inplace=True)
self.backbone_fpn_inner_blocks_3 = torch.nn.Conv2d(1280, 64, kernel_size=(1, 1), stride=(1, 1))
self.backbone_fpn_inner_blocks_2 = torch.nn.Conv2d(112, 64, kernel_size=(1, 1), stride=(1, 1))
self.backbone_fpn_layer_blocks_2_conv_0 = torch.nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=64)
self.backbone_fpn_layer_blocks_2_conv_1 = torch.nn.Conv2d(64, 32, kernel_size=(1, 1), stride=(1, 1))
self.backbone_fpn_layer_blocks_2_conv_2 = torch.nn.ReLU(inplace=True)
self.backbone_fpn_inner_blocks_1 = torch.nn.Conv2d(56, 32, kernel_size=(1, 1), stride=(1, 1))
self.backbone_fpn_layer_blocks_1_conv_0 = torch.nn.Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32)
self.backbone_fpn_layer_blocks_1_conv_1 = torch.nn.Conv2d(32, 24, kernel_size=(1, 1), stride=(1, 1))
self.backbone_fpn_layer_blocks_1_conv_2 = torch.nn.ReLU(inplace=True)
self.backbone_fpn_inner_blocks_0 = torch.nn.Conv2d(40, 24, kernel_size=(1, 1), stride=(1, 1))
self.backbone_fpn_layer_blocks_0_conv_0 = torch.nn.Conv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=24)
self.backbone_fpn_layer_blocks_0_conv_1 = torch.nn.Conv2d(24, 24, kernel_size=(1, 1), stride=(1, 1))
self.backbone_fpn_layer_blocks_0_conv_2 = torch.nn.ReLU(inplace=True)
self.hm_0 = torch.nn.Conv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=24)
self.hm_1 = torch.nn.Conv2d(24, 96, kernel_size=(1, 1), stride=(1, 1))
self.hm_2 = torch.nn.ReLU(inplace=True)
self.hm_3 = torch.nn.Conv2d(96, 1, kernel_size=(1, 1), stride=(1, 1))
self.hps_0 = torch.nn.Conv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=24)
self.hps_1 = torch.nn.Conv2d(24, 96, kernel_size=(1, 1), stride=(1, 1))
self.hps_2 = torch.nn.ReLU(inplace=True)
self.hps_3 = torch.nn.Conv2d(96, 34, kernel_size=(1, 1), stride=(1, 1))
self.hm_hp_0 = torch.nn.Conv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=24)
self.hm_hp_1 = torch.nn.Conv2d(24, 96, kernel_size=(1, 1), stride=(1, 1))
self.hm_hp_2 = torch.nn.ReLU(inplace=True)
self.hm_hp_3 = torch.nn.Conv2d(96, 17, kernel_size=(1, 1), stride=(1, 1))
self.hp_offset_0 = torch.nn.Conv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=24)
self.hp_offset_1 = torch.nn.Conv2d(24, 96, kernel_size=(1, 1), stride=(1, 1))
self.hp_offset_2 = torch.nn.ReLU(inplace=True)
self.hp_offset_3 = torch.nn.Conv2d(96, 34, kernel_size=(1, 1), stride=(1, 1))
self.fake_dequant_0 = torch.quantization.DeQuantStub()
self.fake_dequant_1 = torch.quantization.DeQuantStub()
self.fake_dequant_2 = torch.quantization.DeQuantStub()
self.fake_dequant_3 = torch.quantization.DeQuantStub()
self.float_functional_simple_0 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_1 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_2 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_3 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_4 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_5 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_6 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_7 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_8 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_9 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_10 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_11 = torch.nn.quantized.FloatFunctional()
self.float_functional_simple_12 = torch.nn.quantized.FloatFunctional()
def forward(self, input_1):
fake_quant_0 = self.fake_quant_0(input_1)
backbone_body_0_0 = self.backbone_body_0_0(fake_quant_0)
backbone_body_0_1 = self.backbone_body_0_1(backbone_body_0_0)
backbone_body_0_2 = self.backbone_body_0_2(backbone_body_0_1)
backbone_body_1_conv_0_0 = self.backbone_body_1_conv_0_0(backbone_body_0_2)
backbone_body_1_conv_0_1 = self.backbone_body_1_conv_0_1(backbone_body_1_conv_0_0)
backbone_body_1_conv_1 = self.backbone_body_1_conv_1(backbone_body_1_conv_0_1)
backbone_body_2_conv_0_0 = self.backbone_body_2_conv_0_0(backbone_body_1_conv_1)
backbone_body_2_conv_0_1 = self.backbone_body_2_conv_0_1(backbone_body_2_conv_0_0)
backbone_body_2_conv_1_0 = self.backbone_body_2_conv_1_0(backbone_body_2_conv_0_1)
backbone_body_2_conv_1_1 = self.backbone_body_2_conv_1_1(backbone_body_2_conv_1_0)
backbone_body_2_conv_1_2 = self.backbone_body_2_conv_1_2(backbone_body_2_conv_1_1)
backbone_body_2_conv_2 = self.backbone_body_2_conv_2(backbone_body_2_conv_1_2)
backbone_body_3_conv_0_0 = self.backbone_body_3_conv_0_0(backbone_body_2_conv_2)
backbone_body_3_conv_0_1 = self.backbone_body_3_conv_0_1(backbone_body_3_conv_0_0)
backbone_body_3_conv_1_0 = self.backbone_body_3_conv_1_0(backbone_body_3_conv_0_1)
backbone_body_3_conv_1_1 = self.backbone_body_3_conv_1_1(backbone_body_3_conv_1_0)
backbone_body_3_conv_2 = self.backbone_body_3_conv_2(backbone_body_3_conv_1_1)
add_1 = self.float_functional_simple_0.add(backbone_body_2_conv_2, backbone_body_3_conv_2)
backbone_body_4_conv_0_0 = self.backbone_body_4_conv_0_0(add_1)
backbone_body_4_conv_0_1 = self.backbone_body_4_conv_0_1(backbone_body_4_conv_0_0)
backbone_body_4_conv_1_0 = self.backbone_body_4_conv_1_0(backbone_body_4_conv_0_1)
backbone_body_4_conv_1_1 = self.backbone_body_4_conv_1_1(backbone_body_4_conv_1_0)
backbone_body_4_conv_1_2 = self.backbone_body_4_conv_1_2(backbone_body_4_conv_1_1)
backbone_body_4_conv_2 = self.backbone_body_4_conv_2(backbone_body_4_conv_1_2)
backbone_body_5_conv_0_0 = self.backbone_body_5_conv_0_0(backbone_body_4_conv_2)
backbone_body_5_conv_0_1 = self.backbone_body_5_conv_0_1(backbone_body_5_conv_0_0)
backbone_body_5_conv_1_0 = self.backbone_body_5_conv_1_0(backbone_body_5_conv_0_1)
backbone_body_5_conv_1_1 = self.backbone_body_5_conv_1_1(backbone_body_5_conv_1_0)
backbone_body_5_conv_2 = self.backbone_body_5_conv_2(backbone_body_5_conv_1_1)
add_2 = self.float_functional_simple_1.add(backbone_body_4_conv_2, backbone_body_5_conv_2)
backbone_body_6_conv_0_0 = self.backbone_body_6_conv_0_0(add_2)
backbone_body_6_conv_0_1 = self.backbone_body_6_conv_0_1(backbone_body_6_conv_0_0)
backbone_body_6_conv_1_0 = self.backbone_body_6_conv_1_0(backbone_body_6_conv_0_1)
backbone_body_6_conv_1_1 = self.backbone_body_6_conv_1_1(backbone_body_6_conv_1_0)
backbone_body_6_conv_2 = self.backbone_body_6_conv_2(backbone_body_6_conv_1_1)
add_3 = self.float_functional_simple_2.add(add_2, backbone_body_6_conv_2)
backbone_body_7_conv_0_0 = self.backbone_body_7_conv_0_0(add_3)
backbone_body_7_conv_0_1 = self.backbone_body_7_conv_0_1(backbone_body_7_conv_0_0)
backbone_body_7_conv_1_0 = self.backbone_body_7_conv_1_0(backbone_body_7_conv_0_1)
backbone_body_7_conv_1_1 = self.backbone_body_7_conv_1_1(backbone_body_7_conv_1_0)
backbone_body_7_conv_1_2 = self.backbone_body_7_conv_1_2(backbone_body_7_conv_1_1)
backbone_body_7_conv_2 = self.backbone_body_7_conv_2(backbone_body_7_conv_1_2)
backbone_body_8_conv_0_0 = self.backbone_body_8_conv_0_0(backbone_body_7_conv_2)
backbone_body_8_conv_0_1 = self.backbone_body_8_conv_0_1(backbone_body_8_conv_0_0)
backbone_body_8_conv_1_0 = self.backbone_body_8_conv_1_0(backbone_body_8_conv_0_1)
backbone_body_8_conv_1_1 = self.backbone_body_8_conv_1_1(backbone_body_8_conv_1_0)
backbone_body_8_conv_2 = self.backbone_body_8_conv_2(backbone_body_8_conv_1_1)
add_4 = self.float_functional_simple_3.add(backbone_body_7_conv_2, backbone_body_8_conv_2)
backbone_body_9_conv_0_0 = self.backbone_body_9_conv_0_0(add_4)
backbone_body_9_conv_0_1 = self.backbone_body_9_conv_0_1(backbone_body_9_conv_0_0)
backbone_body_9_conv_1_0 = self.backbone_body_9_conv_1_0(backbone_body_9_conv_0_1)
backbone_body_9_conv_1_1 = self.backbone_body_9_conv_1_1(backbone_body_9_conv_1_0)
backbone_body_9_conv_2 = self.backbone_body_9_conv_2(backbone_body_9_conv_1_1)
add_5 = self.float_functional_simple_4.add(add_4, backbone_body_9_conv_2)
backbone_body_10_conv_0_0 = self.backbone_body_10_conv_0_0(add_5)
backbone_body_10_conv_0_1 = self.backbone_body_10_conv_0_1(backbone_body_10_conv_0_0)
backbone_body_10_conv_1_0 = self.backbone_body_10_conv_1_0(backbone_body_10_conv_0_1)
backbone_body_10_conv_1_1 = self.backbone_body_10_conv_1_1(backbone_body_10_conv_1_0)
backbone_body_10_conv_2 = self.backbone_body_10_conv_2(backbone_body_10_conv_1_1)
add_6 = self.float_functional_simple_5.add(add_5, backbone_body_10_conv_2)
backbone_body_11_conv_0_0 = self.backbone_body_11_conv_0_0(add_6)
backbone_body_11_conv_0_1 = self.backbone_body_11_conv_0_1(backbone_body_11_conv_0_0)
backbone_body_11_conv_1_0 = self.backbone_body_11_conv_1_0(backbone_body_11_conv_0_1)
backbone_body_11_conv_1_1 = self.backbone_body_11_conv_1_1(backbone_body_11_conv_1_0)
backbone_body_11_conv_2 = self.backbone_body_11_conv_2(backbone_body_11_conv_1_1)
backbone_body_12_conv_0_0 = self.backbone_body_12_conv_0_0(backbone_body_11_conv_2)
backbone_body_12_conv_0_1 = self.backbone_body_12_conv_0_1(backbone_body_12_conv_0_0)
backbone_body_12_conv_1_0 = self.backbone_body_12_conv_1_0(backbone_body_12_conv_0_1)
backbone_body_12_conv_1_1 = self.backbone_body_12_conv_1_1(backbone_body_12_conv_1_0)
backbone_body_12_conv_2 = self.backbone_body_12_conv_2(backbone_body_12_conv_1_1)
add_7 = self.float_functional_simple_6.add(backbone_body_11_conv_2, backbone_body_12_conv_2)
backbone_body_13_conv_0_0 = self.backbone_body_13_conv_0_0(add_7)
backbone_body_13_conv_0_1 = self.backbone_body_13_conv_0_1(backbone_body_13_conv_0_0)
backbone_body_13_conv_1_0 = self.backbone_body_13_conv_1_0(backbone_body_13_conv_0_1)
backbone_body_13_conv_1_1 = self.backbone_body_13_conv_1_1(backbone_body_13_conv_1_0)
backbone_body_13_conv_2 = self.backbone_body_13_conv_2(backbone_body_13_conv_1_1)
add_8 = self.float_functional_simple_7.add(add_7, backbone_body_13_conv_2)
backbone_body_14_conv_0_0 = self.backbone_body_14_conv_0_0(add_8)
backbone_body_14_conv_0_1 = self.backbone_body_14_conv_0_1(backbone_body_14_conv_0_0)
backbone_body_14_conv_1_0 = self.backbone_body_14_conv_1_0(backbone_body_14_conv_0_1)
backbone_body_14_conv_1_1 = self.backbone_body_14_conv_1_1(backbone_body_14_conv_1_0)
backbone_body_14_conv_1_2 = self.backbone_body_14_conv_1_2(backbone_body_14_conv_1_1)
backbone_body_14_conv_2 = self.backbone_body_14_conv_2(backbone_body_14_conv_1_2)
backbone_body_15_conv_0_0 = self.backbone_body_15_conv_0_0(backbone_body_14_conv_2)
backbone_body_15_conv_0_1 = self.backbone_body_15_conv_0_1(backbone_body_15_conv_0_0)
backbone_body_15_conv_1_0 = self.backbone_body_15_conv_1_0(backbone_body_15_conv_0_1)
backbone_body_15_conv_1_1 = self.backbone_body_15_conv_1_1(backbone_body_15_conv_1_0)
backbone_body_15_conv_2 = self.backbone_body_15_conv_2(backbone_body_15_conv_1_1)
add_9 = self.float_functional_simple_8.add(backbone_body_14_conv_2, backbone_body_15_conv_2)
backbone_body_16_conv_0_0 = self.backbone_body_16_conv_0_0(add_9)
backbone_body_16_conv_0_1 = self.backbone_body_16_conv_0_1(backbone_body_16_conv_0_0)
backbone_body_16_conv_1_0 = self.backbone_body_16_conv_1_0(backbone_body_16_conv_0_1)
backbone_body_16_conv_1_1 = self.backbone_body_16_conv_1_1(backbone_body_16_conv_1_0)
backbone_body_16_conv_2 = self.backbone_body_16_conv_2(backbone_body_16_conv_1_1)
add_10 = self.float_functional_simple_9.add(add_9, backbone_body_16_conv_2)
backbone_body_17_conv_0_0 = self.backbone_body_17_conv_0_0(add_10)
backbone_body_17_conv_0_1 = self.backbone_body_17_conv_0_1(backbone_body_17_conv_0_0)
backbone_body_17_conv_1_0 = self.backbone_body_17_conv_1_0(backbone_body_17_conv_0_1)
backbone_body_17_conv_1_1 = self.backbone_body_17_conv_1_1(backbone_body_17_conv_1_0)
backbone_body_17_conv_2 = self.backbone_body_17_conv_2(backbone_body_17_conv_1_1)
backbone_body_18_0 = self.backbone_body_18_0(backbone_body_17_conv_2)
backbone_body_18_1 = self.backbone_body_18_1(backbone_body_18_0)
backbone_fpn_inner_blocks_3 = self.backbone_fpn_inner_blocks_3(backbone_body_18_1)
backbone_fpn_inner_blocks_2 = self.backbone_fpn_inner_blocks_2(add_6)
interpolate_1 = torch.nn.functional.interpolate(backbone_fpn_inner_blocks_3, scale_factor=2, mode='bilinear', align_corners=False)
add_11 = self.float_functional_simple_10.add(backbone_fpn_inner_blocks_2, interpolate_1)
backbone_fpn_layer_blocks_2_conv_0 = self.backbone_fpn_layer_blocks_2_conv_0(add_11)
backbone_fpn_layer_blocks_2_conv_1 = self.backbone_fpn_layer_blocks_2_conv_1(backbone_fpn_layer_blocks_2_conv_0)
backbone_fpn_layer_blocks_2_conv_2 = self.backbone_fpn_layer_blocks_2_conv_2(backbone_fpn_layer_blocks_2_conv_1)
backbone_fpn_inner_blocks_1 = self.backbone_fpn_inner_blocks_1(add_3)
interpolate_2 = torch.nn.functional.interpolate(backbone_fpn_layer_blocks_2_conv_2, scale_factor=2, mode='bilinear', align_corners=False)
add_12 = self.float_functional_simple_11.add(backbone_fpn_inner_blocks_1, interpolate_2)
backbone_fpn_layer_blocks_1_conv_0 = self.backbone_fpn_layer_blocks_1_conv_0(add_12)
backbone_fpn_layer_blocks_1_conv_1 = self.backbone_fpn_layer_blocks_1_conv_1(backbone_fpn_layer_blocks_1_conv_0)
backbone_fpn_layer_blocks_1_conv_2 = self.backbone_fpn_layer_blocks_1_conv_2(backbone_fpn_layer_blocks_1_conv_1)
backbone_fpn_inner_blocks_0 = self.backbone_fpn_inner_blocks_0(add_1)
interpolate_3 = torch.nn.functional.interpolate(backbone_fpn_layer_blocks_1_conv_2, scale_factor=2, mode='bilinear', align_corners=False)
add_13 = self.float_functional_simple_12.add(backbone_fpn_inner_blocks_0, interpolate_3)
backbone_fpn_layer_blocks_0_conv_0 = self.backbone_fpn_layer_blocks_0_conv_0(add_13)
backbone_fpn_layer_blocks_0_conv_1 = self.backbone_fpn_layer_blocks_0_conv_1(backbone_fpn_layer_blocks_0_conv_0)
backbone_fpn_layer_blocks_0_conv_2 = self.backbone_fpn_layer_blocks_0_conv_2(backbone_fpn_layer_blocks_0_conv_1)
hm_0 = self.hm_0(backbone_fpn_layer_blocks_0_conv_2)
hm_1 = self.hm_1(hm_0)
hm_2 = self.hm_2(hm_1)
hm_3 = self.hm_3(hm_2)
hps_0 = self.hps_0(backbone_fpn_layer_blocks_0_conv_2)
hps_1 = self.hps_1(hps_0)
hps_2 = self.hps_2(hps_1)
hps_3 = self.hps_3(hps_2)
hm_hp_0 = self.hm_hp_0(backbone_fpn_layer_blocks_0_conv_2)
hm_hp_1 = self.hm_hp_1(hm_hp_0)
hm_hp_2 = self.hm_hp_2(hm_hp_1)
hm_hp_3 = self.hm_hp_3(hm_hp_2)
hp_offset_0 = self.hp_offset_0(backbone_fpn_layer_blocks_0_conv_2)
hp_offset_1 = self.hp_offset_1(hp_offset_0)
hp_offset_2 = self.hp_offset_2(hp_offset_1)
hp_offset_3 = self.hp_offset_3(hp_offset_2)
fake_dequant_0 = self.fake_dequant_0(hm_3)
fake_dequant_1 = self.fake_dequant_1(hps_3)
fake_dequant_2 = self.fake_dequant_2(hm_hp_3)
fake_dequant_3 = self.fake_dequant_3(hp_offset_3)
return fake_dequant_0, fake_dequant_1, fake_dequant_2, fake_dequant_3
if name == "main":
model = MoveNet_qat()
model.load_state_dict(torch.load('out/movenet_qat.pth'))
model.eval()
model.cpu()
dummy_input_0 = torch.ones((1, 3, 256, 256), dtype=torch.float32)
output = model(dummy_input_0)
print(output)
from tinyneuralnetwork.
liamsun2019
commented on May 2, 2024
from tinyneuralnetwork.
peterjc123
commented on May 2, 2024
Thanks for sharing those files with us. But unfortunately, I still cannot reproduce that error locally.
- The device of all the named parameters are cuda:0 as expected
What about adding the check somewhere in the body of the function trainer.train_one_epoch and trainer.validate. As you can see here,
the train function is simply defers the the model to the train_func and validate_func directly, which means that the model won't be modified by our code.
from tinyneuralnetwork.
liamsun2019
commented on May 2, 2024
I find out the root cause now. I need to set explicitly to cuda in train_one_epoch. The run time error disappears after the fix. Thanks for your generous help.
from tinyneuralnetwork.
peterjc123
commented on May 2, 2024
I'm glad you get it fixed.
from tinyneuralnetwork.
Related Issues (20)
- Meet Detailed error: Tensor-likes are not close! using TFLiteConverter HOT 2
- [Converter] Need transpose optimization HOT 2
- Float model failed to convert to TFLite
- [converter] map gather(+reshape) ops with seperate consecutive indices to split(unpack) ops
- tinynn.converter module not found! HOT 2
- [CI] several tests for modifier failed
- Whether to support pytorch to keras HOT 1
- TransposeConv wrong shape? HOT 15
- change input to INT8 after converting to tflite HOT 2
- [converter] implement torch's `aten::scaled_dot_product_attention` operator HOT 2
- Request: clamp would be more efficient to go to Bounded Relu than Maximum + Minimum HOT 3
- Do not support PReLU module? HOT 5
- torch.max not working HOT 2
- OneShotChannelPruner results in the miss of some operators HOT 4
- KeyError when executing quantization HOT 5
- PyTorch 转 TFLite 使用 int8 量化 HOT 4
- Does tinynn support following int16 quantization? HOT 1
- jit.trace succeed but tinynn tracer failed HOT 1
- It became larger after converting to tflite model HOT 4
- how to do Post-training integer quantization with int16 activation HOT 4
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from tinyneuralnetwork.