micronet, a model compression and deploy lib. compression: 1、quantization: quantization-aware-training(QAT), High-Bit(>2b)(DoReFa/Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference)、Low-Bit(≤2b)/Ternary and Binary(TWN/BNN/XNOR-Net); post-training-quantization(PTQ), 8-bit(tensorrt); 2、 pruning: normal、regular and group convolutional channel pruning; 3、 group convolution structure; 4、batch-normalization fuse for quantization. deploy: tensorrt, fp32/fp16/int8(ptq-calibration)、op-adapt(upsample)、dynamic_shape
250zhanghu
commented
2 years ago
(pytorch1) maziyi@kpyf:~/python/yolox5/micronet/micronet/compression/quantization/wqaq/iao$ python main.py --resume "/home/maziyi/python/yolox5/yolox_best.pth" --q_type 0 --q_level 0 --bn_fuse --qaft --lr 0.00001 ==> Options: Namespace(a_bits=8, batch_size=32, bn_fuse=True, bn_fuse_calib=False, cpu=False, data='../../../../data', end_epochs=300, eval_batch_size=32, gpu_id='', lr='0.00001', model_type=1, num_workers=2, percentile=0.999999, pretrained_model=False, prune_qaft='', prune_quant='', ptq=False, ptq_batch=200, ptq_control=False, q_level=0, q_type=0, qaft=True, refine='', resume='/home/maziyi/python/yolox5/yolox_best.pth', start_epochs=1, w_bits=8, wd=1e-05, weight_observer=0) ==> Preparing data.. Files already downloaded and verified Files already downloaded and verified **Reume model** ori_model Net( (model): Sequential( (0): ConvBNReLU( (conv): Conv2d(3, 256, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2)) (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (1): ConvBNReLU( (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), groups=2) (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (2): ConvBNReLU( (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), groups=2) (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) (4): ConvBNReLU( (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=16) (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (5): ConvBNReLU( (conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), groups=4) (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (6): ConvBNReLU( (conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), groups=4) (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (7): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) (8): ConvBNReLU( (conv): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32) (bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (9): ConvBNReLU( (conv): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1), groups=8) (bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (10): ConvBNReLU( (conv): Conv2d(1024, 10, kernel_size=(1, 1), stride=(1, 1)) (bn): BatchNorm2d(10, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (relu): ReLU(inplace=True) ) (11): AvgPool2d(kernel_size=8, stride=1, padding=0) ) )
quant_model Net( (model): Sequential( (0): ConvBNReLU( (conv): QuantBNFuseConv2d( 3, 256, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2) (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (1): ConvBNReLU( (conv): QuantBNFuseConv2d( 256, 256, kernel_size=(1, 1), stride=(1, 1), groups=2 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (2): ConvBNReLU( (conv): QuantBNFuseConv2d( 256, 256, kernel_size=(1, 1), stride=(1, 1), groups=2 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (3): QuantMaxPool2d( kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) ) (4): ConvBNReLU( (conv): QuantBNFuseConv2d( 256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=16 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (5): ConvBNReLU( (conv): QuantBNFuseConv2d( 512, 512, kernel_size=(1, 1), stride=(1, 1), groups=4 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (6): ConvBNReLU( (conv): QuantBNFuseConv2d( 512, 512, kernel_size=(1, 1), stride=(1, 1), groups=4 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (7): QuantMaxPool2d( kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) ) (8): ConvBNReLU( (conv): QuantBNFuseConv2d( 512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (9): ConvBNReLU( (conv): QuantBNFuseConv2d( 1024, 1024, kernel_size=(1, 1), stride=(1, 1), groups=8 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (10): ConvBNReLU( (conv): QuantBNFuseConv2d( 1024, 10, kernel_size=(1, 1), stride=(1, 1) (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) (weight_quantizer): SymmetricQuantizer( (observer): MinMaxObserver() ) ) (bn): Identity() (relu): ReLU(inplace=True) ) (11): QuantAvgPool2d( kernel_size=8, stride=1, padding=0 (activation_quantizer): SymmetricQuantizer( (observer): MovingAverageMinMaxObserver() ) ) ) ) Traceback (most recent call last): File "main.py", line 479, in
model.load_state_dict(checkpoint["state_dict"])
KeyError: 'state_dict'
求大佬解答