Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

quantized.out 量化OCR模型出错 #3154

Open
Tmn07 opened this issue Jan 7, 2025 · 5 comments
Open

quantized.out 量化OCR模型出错 #3154

Tmn07 opened this issue Jan 7, 2025 · 5 comments
Labels
quantization

Comments

@Tmn07
Copy link

Tmn07 commented Jan 7, 2025

模型来自https://paddlepaddle.github.io/PaddleOCR/main/ppocr/model_list.html#21
ch_PP-OCRv3_rec

现将paddle模型转为onnx，再转为fp32 mnn，再进行量化
以rec模型为例

paddle2onnx --model_dir ./inference/ch_PP-OCRv3_rec_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_file ./inference/rec_onnx/model.onnx \
--opset_version 11 \
--enable_onnx_checker True

./MNNConvert -f ONNX --modelFile rec.onnx --MNNModel rec.mnn --bizCode biz

./quantized.out rec.mnn rec-quant.mnn imageInputConfig.json

运行报错
Compute Shape Error for x___tr4conv2d_97.tmp_0，其中x___tr4conv2d_97.tmp_0像是网络结构中第一层的一个shape转换NCHW->NC4HW4。这个ConvertTensor层无法正常推理？有什么思路来进一步定位和解决问题呢

日志输出Quantize model done!，但实际输出模型大小还是32位，并不是指定的8bit

imageInputConfig.json

{
    "format":"RGB",
    "mean":[
        127.5,
        127.5,
        127.5
    ],
    "normal":[
        0.00784314,
        0.00784314,
        0.00784314
    ],
    "width":48,
    "height": 320,
    "center_crop_h":0.875,
    "center_crop_w":0.875,
    "path":"/home/vca1/swq/data/ocr/calib/48x320",
    "used_image_num":1,
    "feature_quantize_method":"KL",
    "weight_quantize_method":"MAX_ABS",
    "batch_size":1,
    "quant_bits":8,
    "input_type":"image",
    "debug":true
}

模型文件
rec.onnx.zip

rec.mnn.zip

完整运行日志

[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1465: >>> modelFile: /home/vca1/swq/code/PaddleOCR/models/zwd_mnn_fp32_new/rec.mnn
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1466: >>> preTreatConfig: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/imageInputConfig.json
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1467: >>> dstFile: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/rec-quant.mnn
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1495: Calibrate the feature and quantize model...
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:306: Use feature quantization method: KL
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:307: Use weight quantization method: MAX_ABS
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:327: feature_clamp_value: 127
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:328: weight_clamp_value: 127
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:332: winogradOpt only be available under EMA
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/Helper.cpp:143: used dataset num: 1
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:830: fake quant weights done.
Invalid Tensor, the session may not be ready
Compute Shape Error for x___tr4conv2d_97.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for x___tr4conv2d_97.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for x___tr4conv2d_97.tmp_0
code=3 in onForward, 564 
CollectFeatureDistribution: 100.00 %
Invalid Tensor, the session may not be ready
Compute Shape Error for x___tr4conv2d_97.tmp_0
code=3 in onForward, 564 
Compute Shape Error for x___tr4conv2d_97.tmp_0
code=3 in onForward, 564 
computeDistance: 100.00 %

Debug info:

Can't find input extraTensorDescribe for batch_norm_27.tmp_4
Can't find input extraTensorDescribe for batch_norm_28.tmp_4
Can't find input extraTensorDescribe for batch_norm_29.tmp_4
Can't find input extraTensorDescribe for batch_norm_30.tmp_4
Can't find input extraTensorDescribe for batch_norm_31.tmp_4
Can't find input extraTensorDescribe for batch_norm_32.tmp_4
Can't find input extraTensorDescribe for batch_norm_33.tmp_4
Can't find input extraTensorDescribe for batch_norm_34.tmp_4
Can't find input extraTensorDescribe for batch_norm_35.tmp_4
Can't find input extraTensorDescribe for batch_norm_36.tmp_4
Can't find input extraTensorDescribe for batch_norm_37.tmp_4
Can't find input extraTensorDescribe for batch_norm_38.tmp_4
Can't find input extraTensorDescribe for batch_norm_39.tmp_4
Can't find input extraTensorDescribe for batch_norm_40.tmp_4
Can't find input extraTensorDescribe for batch_norm_41.tmp_4
Can't find input extraTensorDescribe for batch_norm_42.tmp_4
Can't find input extraTensorDescribe for batch_norm_43.tmp_4
Can't find input extraTensorDescribe for batch_norm_44.tmp_4
Can't find input extraTensorDescribe for batch_norm_45.tmp_4
Can't find input extraTensorDescribe for batch_norm_46.tmp_4
Can't find input extraTensorDescribe for batch_norm_47.tmp_4
Can't find input extraTensorDescribe for batch_norm_48.tmp_4
Can't find input extraTensorDescribe for batch_norm_49.tmp_4
Can't find input extraTensorDescribe for batch_norm_50.tmp_4
Can't find input extraTensorDescribe for batch_norm_51.tmp_4
Can't find input extraTensorDescribe for batch_norm_52.tmp_4
Can't find input extraTensorDescribe for batch_norm_53.tmp_4
Can't find input extraTensorDescribe for p2o.Sigmoid.1
Can't find input extraTensorDescribe for p2o.Sigmoid.3
Can't find input extraTensorDescribe for p2o.Sigmoid.5
Can't find input extraTensorDescribe for p2o.Sigmoid.7
Can't find input extraTensorDescribe for p2o.Sigmoid.9
Can't find input extraTensorDescribe for p2o.Sigmoid.11
Can't find input extraTensorDescribe for p2o.Sigmoid.13
[15:51:02] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1503: Quantize model done!

平台(如果交叉编译请再附上交叉编译目标平台):

Platform(Include target platform as well if cross-compiling):

linux

Github版本:

Github Version:

3.0

编译方式:

Compiling Method

 cmake ..  -DMNN_BUILD_CONVERTER=ON -DMNN_BUILD_QUANTOOLS=ON -DMNN_LOW_MEMORY=ON
make -j8
@jxt1234
Copy link
Collaborator

jxt1234 commented Jan 7, 2025

转 mnn 时加个 --keepInputFormat=0 试试？
./MNNConvert -f ONNX --modelFile rec.onnx --MNNModel rec.mnn --bizCode biz --keepInputFormat=0

@Tmn07
Copy link
Author

Tmn07 commented Jan 7, 2025

@jxt1234 感谢回复!
尝试加了--keepInputFormat=0 ，第一个层的ConvertTensor确实没了。但量化时后续卷积层又报错了，是哪里用的不对吗

./quantized.out rec.mnn rec-quant.mnn imageInputConfig.json

运行日志如下

[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1465: >>> modelFile: /home/vca1/swq/code/PaddleOCR/models/zwd_mnn_fp32_new/rec.mnn
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1466: >>> preTreatConfig: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/imageInputConfig.json
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1467: >>> dstFile: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/rec-quant.mnn
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1495: Calibrate the feature and quantize model...
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:306: Use feature quantization method: KL
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:307: Use weight quantization method: MAX_ABS
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:327: feature_clamp_value: 127
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:328: weight_clamp_value: 127
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:332: winogradOpt only be available under EMA
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/Helper.cpp:143: used dataset num: 1
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:830: fake quant weights done.
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_97.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_97.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_97.tmp_0
code=3 in onForward, 564 
CollectFeatureDistribution: 100.00 %
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_97.tmp_0
code=3 in onForward, 564 
Compute Shape Error for conv2d_97.tmp_0
code=3 in onForward, 564 
computeDistance: 100.00 %

Debug info:

Can't find input extraTensorDescribe for batch_norm_27.tmp_4
Can't find input extraTensorDescribe for batch_norm_28.tmp_4
Can't find input extraTensorDescribe for batch_norm_29.tmp_4
Can't find input extraTensorDescribe for batch_norm_30.tmp_4
Can't find input extraTensorDescribe for batch_norm_31.tmp_4
Can't find input extraTensorDescribe for batch_norm_32.tmp_4
Can't find input extraTensorDescribe for batch_norm_33.tmp_4
Can't find input extraTensorDescribe for batch_norm_34.tmp_4
Can't find input extraTensorDescribe for batch_norm_35.tmp_4
Can't find input extraTensorDescribe for batch_norm_36.tmp_4
Can't find input extraTensorDescribe for batch_norm_37.tmp_4
Can't find input extraTensorDescribe for batch_norm_38.tmp_4
Can't find input extraTensorDescribe for batch_norm_39.tmp_4
Can't find input extraTensorDescribe for batch_norm_40.tmp_4
Can't find input extraTensorDescribe for batch_norm_41.tmp_4
Can't find input extraTensorDescribe for batch_norm_42.tmp_4
Can't find input extraTensorDescribe for batch_norm_43.tmp_4
Can't find input extraTensorDescribe for batch_norm_44.tmp_4
Can't find input extraTensorDescribe for batch_norm_45.tmp_4
Can't find input extraTensorDescribe for batch_norm_46.tmp_4
Can't find input extraTensorDescribe for batch_norm_47.tmp_4
Can't find input extraTensorDescribe for batch_norm_48.tmp_4
Can't find input extraTensorDescribe for batch_norm_49.tmp_4
Can't find input extraTensorDescribe for batch_norm_50.tmp_4
Can't find input extraTensorDescribe for batch_norm_51.tmp_4
Can't find input extraTensorDescribe for batch_norm_52.tmp_4
Can't find input extraTensorDescribe for batch_norm_53.tmp_4
Can't find input extraTensorDescribe for swish_21.tmp_0
Can't find input extraTensorDescribe for swish_22.tmp_0
Can't find input extraTensorDescribe for swish_23.tmp_0
Can't find input extraTensorDescribe for swish_24.tmp_0
Can't find input extraTensorDescribe for swish_25.tmp_0
Can't find input extraTensorDescribe for swish_26.tmp_0
Can't find input extraTensorDescribe for swish_27.tmp_0
[20:49:01] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1503: Quantize model done!

@jxt1234
Copy link
Collaborator

jxt1234 commented Jan 8, 2025

rec.mnn 可以直接运行么？比如使用 MNNV2Basic 工具测试?

@Tmn07
Copy link
Author

Tmn07 commented Jan 8, 2025

@jxt1234

用MNNV2Basic工具测试报错了，我平时是用mnn的python接口可以进行推理。

./GetMNNInfo rec.mnn 
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
Model default dimensionFormat is NCHW
Model Inputs:
[ x ]: dimensionFormat: NC4HW4, size: [ 1,3,48,-1 ], type is float
Model Outputs:
[ softmax_5.tmp_0 ]
Model Version: 3.0.1 
Model bizCode: biz

./MNNV2Basic.out rec.mnn 1 0 0 4 2 1x3x48x320
Use extra forward type: 0
1 3 48 320 
Open Model /rec.mnn
Can't open file:.tempcache
Load Cache file error.
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
===========> Resize Again...
ERROR: Unary Op can not execute
ERROR: Unary Op can not execute
Create execution error : 101
test_main, 285, cost time: 30.821001 ms
Resize error, can't execute MNN

另外有一个模型，MNNV2Basic测试可以通过，但量化还是报错

(base) root@101vector:/home/vca1/zwd/MNN-3.0.0/build# ./MNNV2Basic.out cls.mnn 1 1 0 4 2 1x3x23x23
Save AllTensors to output/*.txt
Use extra forward type: 0
1 3 23 23 
Open Model cls.mnn
Can't open file:.tempcache
Load Cache file error.
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
===========> Resize Again...
test_main, 285, cost time: 14.305000 ms
Session Info: memory use 0.587383 MB, flops is 1.246171 M, backendType is 13
===========> Session Resize Done.
===========> Session Start running...
Input size:2116
	**Tensor shape**: 1, 3, 23, 23, 
fileName.str().c_str()=s ./input_0.txt in _loadInputFromFile, 110 
Dimensions: 4, W,H,C,B: 12 X 12 X 8 X 1, OP name conv2d_53.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 12 X 8 X 1, OP name hardswish_0.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 12 X 8 X 1, OP name conv2d_54.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name depthwise_conv2d_0.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name p2o.GlobalAveragePool.1_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 2 X 1, OP name conv2d_55.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name conv2d_56.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name BinaryOp21 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name BinaryOp23 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name hardsigmoid_0.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name p2o.Mul.3_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name p2o.Mul.3 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name conv2d_57.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 24 X 1, OP name conv2d_58.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 24 X 1, OP name depthwise_conv2d_1.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 8 X 1, OP name conv2d_59.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name conv2d_60.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name depthwise_conv2d_2.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 8 X 1, OP name conv2d_61.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 8 X 1, OP name p2o.Add.7 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name conv2d_62.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name hardswish_1.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name depthwise_conv2d_3.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name hardswish_2.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name p2o.GlobalAveragePool.3_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name conv2d_63.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name conv2d_64.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name BinaryOp62 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name BinaryOp64 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name hardsigmoid_1.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name p2o.Mul.9_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name p2o.Mul.9 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_65.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name conv2d_66.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_3.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name depthwise_conv2d_4.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_4.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name p2o.GlobalAveragePool.5_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 22 X 1, OP name conv2d_67.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name conv2d_68.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp86 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp88 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name hardsigmoid_2.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.15_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.15 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_69.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.25 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name conv2d_70.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_5.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name depthwise_conv2d_5.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_6.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name p2o.GlobalAveragePool.7_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 22 X 1, OP name conv2d_71.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name conv2d_72.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp111 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp113 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name hardsigmoid_3.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.21_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.21 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_73.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.35 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name conv2d_74.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name hardswish_7.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name depthwise_conv2d_6.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name hardswish_8.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name p2o.GlobalAveragePool.9_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 10 X 1, OP name conv2d_75.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name conv2d_76.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name BinaryOp136 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name BinaryOp138 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name hardsigmoid_4.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name p2o.Mul.27_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name p2o.Mul.27 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_77.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.45 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name conv2d_78.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name hardswish_9.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name depthwise_conv2d_7.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name hardswish_10.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name p2o.GlobalAveragePool.11_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 12 X 1, OP name conv2d_79.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name conv2d_80.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name BinaryOp161 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name BinaryOp163 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name hardsigmoid_5.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name p2o.Mul.33_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name p2o.Mul.33 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_81.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.55 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 104 X 1, OP name conv2d_82.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 104 X 1, OP name hardswish_11.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name depthwise_conv2d_8.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name hardswish_12.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name p2o.GlobalAveragePool.13_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 26 X 1, OP name conv2d_83.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name conv2d_84.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name BinaryOp186 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name BinaryOp188 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name hardsigmoid_6.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name p2o.Mul.39_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name p2o.Mul.39 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name conv2d_85.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name conv2d_86.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_13.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name depthwise_conv2d_9.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_14.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.GlobalAveragePool.15_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 50 X 1, OP name conv2d_87.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name conv2d_88.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp210 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp212 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name hardsigmoid_7.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.45_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.45 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name conv2d_89.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name p2o.Add.73 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name conv2d_90.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_15.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name depthwise_conv2d_10.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_16.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.GlobalAveragePool.17_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 50 X 1, OP name conv2d_91.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name conv2d_92.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp235 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp237 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name hardsigmoid_8.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.51_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.51 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name conv2d_93.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name p2o.Add.83 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name conv2d_94.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_17.tmp_0 : 0
Dimensions: 4, W,H,C,B: 6 X 1 X 200 X 1, OP name p2o.MaxPool.1 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.GlobalAveragePool.19_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.Add.87__matmul_converted_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 2 X 1, OP name p2o.Add.87__matmul_converted : 0
Dimensions: 2, 1 X 2, OP name Softmax264_raster_0 : 0
Dimensions: 2, 1 X 2, OP name Softmax264 : 0
Dimensions: 2, 1 X 2, OP name softmax_0.tmp_0_raster_0 : 0
output: softmax_0.tmp_0
precision:2, memory: 0, Run 1 time:
                          softmax_0.tmp_0_raster_0 	[Raster] run 1 average cost 0.000000 ms, 0.000 %, FlopsRate: 0.000 %
                                        BinaryOp21 	[BinaryOp] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.001 %
                                        BinaryOp23 	[BinaryOp] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.001 %
                                        Softmax264 	[Softmax] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.000 %
                               hardsigmoid_1.tmp_0 	[ReLU6] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.002 %
                                       BinaryOp111 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.007 %
                                       BinaryOp113 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.007 %
                                       BinaryOp136 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.003 %
                                       BinaryOp161 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.004 %
                                       BinaryOp163 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.004 %
                                       BinaryOp212 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.015 %
                               Softmax264_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.000 %
                                   conv2d_55.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.001 %
                                   conv2d_56.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.001 %
                                   conv2d_57.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.370 %
                                   conv2d_79.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.046 %
                          depthwise_conv2d_0.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.416 %
                               hardsigmoid_3.tmp_0 	[ReLU6] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.007 %
                               hardsigmoid_7.tmp_0 	[ReLU6] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.015 %
                                hardswish_10.tmp_0 	[UnaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.088 %
                                        p2o.Add.25 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.029 %
                                        p2o.Add.35 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.029 %
                                        p2o.Add.45 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.029 %
                      p2o.Add.87__matmul_converted 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.032 %
                               p2o.Mul.15_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.162 %
                               p2o.Mul.39_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.096 %
                               p2o.Mul.45_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.184 %
                               p2o.Mul.51_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.184 %
                                p2o.Mul.9_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.059 %
                                       BinaryOp138 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.003 %
                                       BinaryOp186 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.008 %
                                       BinaryOp210 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                       BinaryOp235 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                       BinaryOp237 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                        BinaryOp62 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.002 %
                                        BinaryOp64 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.002 %
                                        BinaryOp86 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.007 %
                                        BinaryOp88 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.007 %
                                   conv2d_59.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.555 %
                                   conv2d_63.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.021 %
                                   conv2d_68.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.155 %
                                   conv2d_72.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.155 %
                                   conv2d_75.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.032 %
                                   conv2d_76.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.032 %
                                   conv2d_80.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.046 %
                                   conv2d_83.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.217 %
                                   conv2d_88.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.802 %
                               hardsigmoid_0.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.001 %
                               hardsigmoid_2.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.007 %
                               hardsigmoid_4.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.003 %
                               hardsigmoid_5.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.004 %
                               hardsigmoid_6.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.008 %
                               hardsigmoid_8.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                 hardswish_1.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.088 %
                                hardswish_12.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.096 %
                                hardswish_13.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_14.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_15.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_16.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_17.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                 hardswish_2.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.059 %
                                 hardswish_8.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.073 %
                                        p2o.Add.55 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.029 %
                                         p2o.Add.7 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.022 %
                                        p2o.Add.83 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.029 %
                 p2o.GlobalAveragePool.11_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                 p2o.GlobalAveragePool.13_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.3_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.9_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                                        p2o.Mul.15 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.162 %
                               p2o.Mul.21_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.162 %
                                        p2o.Mul.27 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.073 %
                               p2o.Mul.27_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.073 %
                                         p2o.Mul.3 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.044 %
                               p2o.Mul.33_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.088 %
                                p2o.Mul.3_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.044 %
                                        p2o.Mul.45 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                   conv2d_54.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.740 %
                                   conv2d_61.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.740 %
                                   conv2d_62.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.740 %
                                   conv2d_64.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.021 %
                                   conv2d_65.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.986 %
                                   conv2d_67.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.155 %
                                   conv2d_71.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.155 %
                                   conv2d_77.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 1.233 %
                                   conv2d_78.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 1.479 %
                                   conv2d_84.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.217 %
                                   conv2d_87.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.802 %
                                   conv2d_91.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.802 %
                                   conv2d_92.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.802 %
                          depthwise_conv2d_1.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.624 %
                                 hardswish_4.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.162 %
                                 hardswish_6.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.162 %
                                 hardswish_7.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.073 %
                                 hardswish_9.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.088 %
                 p2o.GlobalAveragePool.19_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.1_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.5_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.7_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                                        p2o.Mul.21 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.162 %
                                        p2o.Mul.33 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.088 %
                                        p2o.Mul.39 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.096 %
                                        p2o.Mul.51 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.184 %
                                         p2o.Mul.9 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.059 %
                                   conv2d_58.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 1.109 %
                                   conv2d_60.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.740 %
                                   conv2d_81.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 1.479 %
                                   conv2d_82.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 3.205 %
                          depthwise_conv2d_2.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.832 %
                          depthwise_conv2d_3.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 1.541 %
                          depthwise_conv2d_7.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 2.311 %
                          depthwise_conv2d_8.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 2.504 %
                                 hardswish_0.tmp_0 	[UnaryOp] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.088 %
                                        p2o.Add.73 	[BinaryOp] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.029 %
             p2o.Add.87__matmul_converted_raster_0 	[Raster] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.015 %
                 p2o.GlobalAveragePool.15_raster_0 	[Pooling] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.000 %
                 p2o.GlobalAveragePool.17_raster_0 	[Pooling] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.000 %
                                     p2o.MaxPool.1 	[Pooling] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.367 %
                                       BinaryOp188 	[BinaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.008 %
                                   conv2d_66.tmp_0 	[Convolution] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 2.712 %
                                   conv2d_70.tmp_0 	[Convolution] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 2.712 %
                                   conv2d_73.tmp_0 	[Convolution] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 2.712 %
                          depthwise_conv2d_5.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 4.237 %
                          depthwise_conv2d_6.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 1.926 %
                                hardswish_11.tmp_0 	[UnaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.191 %
                                 hardswish_3.tmp_0 	[UnaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.162 %
                                 hardswish_5.tmp_0 	[UnaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.162 %
                                   conv2d_69.tmp_0 	[Convolution] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 2.712 %
                                   conv2d_74.tmp_0 	[Convolution] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 1.233 %
                                   conv2d_90.tmp_0 	[Convolution] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 6.163 %
                          depthwise_conv2d_4.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 4.237 %
                                   conv2d_86.tmp_0 	[Convolution] run 1 average cost 0.008000 ms, 1.035 %, FlopsRate: 6.163 %
                                   conv2d_94.tmp_0 	[Convolution] run 1 average cost 0.008000 ms, 1.035 %, FlopsRate: 6.163 %
                          depthwise_conv2d_9.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.008000 ms, 1.035 %, FlopsRate: 4.815 %
                         depthwise_conv2d_10.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.009000 ms, 1.164 %, FlopsRate: 4.815 %
                                   conv2d_85.tmp_0 	[Convolution] run 1 average cost 0.011000 ms, 1.423 %, FlopsRate: 3.205 %
                                   conv2d_93.tmp_0 	[Convolution] run 1 average cost 0.018000 ms, 2.329 %, FlopsRate: 6.163 %
                                   conv2d_89.tmp_0 	[Convolution] run 1 average cost 0.019000 ms, 2.458 %, FlopsRate: 6.163 %
                                   conv2d_53.tmp_0 	[Convolution] run 1 average cost 0.030000 ms, 3.881 %, FlopsRate: 2.496 %
Avg= 0.773000 ms, OpSum = 0.567000 ms min= 0.773000 ms, max= 0.773000 ms

 
./quantized.out cls.mnn cls-quant.mnn imageInputConfig.json
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1465: >>> modelFile: /home/vca1/swq/code/PaddleOCR/models/zwd_mnn_fp32_new/cls.mnn
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1466: >>> preTreatConfig: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/imageInputConfig.json
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1467: >>> dstFile: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/cls-quant.mnn
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1495: Calibrate the feature and quantize model...
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:306: Use feature quantization method: KL
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:307: Use weight quantization method: MAX_ABS
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:327: feature_clamp_value: 127
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:328: weight_clamp_value: 127
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:332: winogradOpt only be available under EMA
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/Helper.cpp:143: used dataset num: 1
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:830: fake quant weights done.
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
CollectFeatureDistribution: 100.00 %
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
computeDistance: 100.00 %

Debug info:

Can't find input extraTensorDescribe for hardswish_0.tmp_0
Can't find input extraTensorDescribe for hardswish_1.tmp_0
Can't find input extraTensorDescribe for hardswish_2.tmp_0
Can't find input extraTensorDescribe for hardswish_3.tmp_0
Can't find input extraTensorDescribe for hardswish_4.tmp_0
Can't find input extraTensorDescribe for hardswish_5.tmp_0
Can't find input extraTensorDescribe for hardswish_6.tmp_0
Can't find input extraTensorDescribe for hardswish_7.tmp_0
Can't find input extraTensorDescribe for hardswish_8.tmp_0
Can't find input extraTensorDescribe for hardswish_9.tmp_0
Can't find input extraTensorDescribe for hardswish_10.tmp_0
Can't find input extraTensorDescribe for hardswish_11.tmp_0
Can't find input extraTensorDescribe for hardswish_12.tmp_0
Can't find input extraTensorDescribe for hardswish_13.tmp_0
Can't find input extraTensorDescribe for hardswish_14.tmp_0
Can't find input extraTensorDescribe for hardswish_15.tmp_0
Can't find input extraTensorDescribe for hardswish_16.tmp_0
Can't find input extraTensorDescribe for hardswish_17.tmp_0
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1503: Quantize model done!

@jxt1234
Copy link
Collaborator

jxt1234 commented Jan 9, 2025

@jxt1234

用MNNV2Basic工具测试报错了，我平时是用mnn的python接口可以进行推理。

./GetMNNInfo rec.mnn 
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
Model default dimensionFormat is NCHW
Model Inputs:
[ x ]: dimensionFormat: NC4HW4, size: [ 1,3,48,-1 ], type is float
Model Outputs:
[ softmax_5.tmp_0 ]
Model Version: 3.0.1 
Model bizCode: biz

./MNNV2Basic.out rec.mnn 1 0 0 4 2 1x3x48x320
Use extra forward type: 0
1 3 48 320 
Open Model /rec.mnn
Can't open file:.tempcache
Load Cache file error.
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
===========> Resize Again...
ERROR: Unary Op can not execute
ERROR: Unary Op can not execute
Create execution error : 101
test_main, 285, cost time: 30.821001 ms
Resize error, can't execute MNN

另外有一个模型，MNNV2Basic测试可以通过，但量化还是报错

(base) root@101vector:/home/vca1/zwd/MNN-3.0.0/build# ./MNNV2Basic.out cls.mnn 1 1 0 4 2 1x3x23x23
Save AllTensors to output/*.txt
Use extra forward type: 0
1 3 23 23 
Open Model cls.mnn
Can't open file:.tempcache
Load Cache file error.
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
===========> Resize Again...
test_main, 285, cost time: 14.305000 ms
Session Info: memory use 0.587383 MB, flops is 1.246171 M, backendType is 13
===========> Session Resize Done.
===========> Session Start running...
Input size:2116
	**Tensor shape**: 1, 3, 23, 23, 
fileName.str().c_str()=s ./input_0.txt in _loadInputFromFile, 110 
Dimensions: 4, W,H,C,B: 12 X 12 X 8 X 1, OP name conv2d_53.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 12 X 8 X 1, OP name hardswish_0.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 12 X 8 X 1, OP name conv2d_54.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name depthwise_conv2d_0.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name p2o.GlobalAveragePool.1_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 2 X 1, OP name conv2d_55.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name conv2d_56.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name BinaryOp21 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name BinaryOp23 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name hardsigmoid_0.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name p2o.Mul.3_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name p2o.Mul.3 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 8 X 1, OP name conv2d_57.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 6 X 24 X 1, OP name conv2d_58.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 24 X 1, OP name depthwise_conv2d_1.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 8 X 1, OP name conv2d_59.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name conv2d_60.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name depthwise_conv2d_2.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 8 X 1, OP name conv2d_61.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 8 X 1, OP name p2o.Add.7 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name conv2d_62.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 3 X 32 X 1, OP name hardswish_1.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name depthwise_conv2d_3.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name hardswish_2.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name p2o.GlobalAveragePool.3_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 8 X 1, OP name conv2d_63.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name conv2d_64.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name BinaryOp62 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name BinaryOp64 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 32 X 1, OP name hardsigmoid_1.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name p2o.Mul.9_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 32 X 1, OP name p2o.Mul.9 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_65.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name conv2d_66.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_3.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name depthwise_conv2d_4.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_4.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name p2o.GlobalAveragePool.5_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 22 X 1, OP name conv2d_67.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name conv2d_68.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp86 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp88 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name hardsigmoid_2.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.15_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.15 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_69.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.25 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name conv2d_70.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_5.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name depthwise_conv2d_5.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name hardswish_6.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name p2o.GlobalAveragePool.7_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 22 X 1, OP name conv2d_71.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name conv2d_72.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp111 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name BinaryOp113 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 88 X 1, OP name hardsigmoid_3.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.21_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 88 X 1, OP name p2o.Mul.21 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_73.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.35 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name conv2d_74.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name hardswish_7.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name depthwise_conv2d_6.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name hardswish_8.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name p2o.GlobalAveragePool.9_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 10 X 1, OP name conv2d_75.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name conv2d_76.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name BinaryOp136 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name BinaryOp138 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 40 X 1, OP name hardsigmoid_4.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name p2o.Mul.27_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 40 X 1, OP name p2o.Mul.27 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_77.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.45 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name conv2d_78.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name hardswish_9.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name depthwise_conv2d_7.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name hardswish_10.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name p2o.GlobalAveragePool.11_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 12 X 1, OP name conv2d_79.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name conv2d_80.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name BinaryOp161 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name BinaryOp163 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 48 X 1, OP name hardsigmoid_5.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name p2o.Mul.33_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 48 X 1, OP name p2o.Mul.33 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name conv2d_81.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 16 X 1, OP name p2o.Add.55 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 104 X 1, OP name conv2d_82.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 2 X 104 X 1, OP name hardswish_11.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name depthwise_conv2d_8.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name hardswish_12.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name p2o.GlobalAveragePool.13_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 26 X 1, OP name conv2d_83.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name conv2d_84.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name BinaryOp186 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name BinaryOp188 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 104 X 1, OP name hardsigmoid_6.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name p2o.Mul.39_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 104 X 1, OP name p2o.Mul.39 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name conv2d_85.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name conv2d_86.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_13.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name depthwise_conv2d_9.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_14.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.GlobalAveragePool.15_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 50 X 1, OP name conv2d_87.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name conv2d_88.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp210 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp212 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name hardsigmoid_7.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.45_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.45 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name conv2d_89.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name p2o.Add.73 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name conv2d_90.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_15.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name depthwise_conv2d_10.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_16.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.GlobalAveragePool.17_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 50 X 1, OP name conv2d_91.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name conv2d_92.tmp_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp235 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name BinaryOp237 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name hardsigmoid_8.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.51_raster_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name p2o.Mul.51 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name conv2d_93.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 32 X 1, OP name p2o.Add.83 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name conv2d_94.tmp_0 : 0
Dimensions: 4, W,H,C,B: 12 X 1 X 200 X 1, OP name hardswish_17.tmp_0 : 0
Dimensions: 4, W,H,C,B: 6 X 1 X 200 X 1, OP name p2o.MaxPool.1 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.GlobalAveragePool.19_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 200 X 1, OP name p2o.Add.87__matmul_converted_raster_0 : 0
Dimensions: 4, W,H,C,B: 1 X 1 X 2 X 1, OP name p2o.Add.87__matmul_converted : 0
Dimensions: 2, 1 X 2, OP name Softmax264_raster_0 : 0
Dimensions: 2, 1 X 2, OP name Softmax264 : 0
Dimensions: 2, 1 X 2, OP name softmax_0.tmp_0_raster_0 : 0
output: softmax_0.tmp_0
precision:2, memory: 0, Run 1 time:
                          softmax_0.tmp_0_raster_0 	[Raster] run 1 average cost 0.000000 ms, 0.000 %, FlopsRate: 0.000 %
                                        BinaryOp21 	[BinaryOp] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.001 %
                                        BinaryOp23 	[BinaryOp] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.001 %
                                        Softmax264 	[Softmax] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.000 %
                               hardsigmoid_1.tmp_0 	[ReLU6] run 1 average cost 0.001000 ms, 0.129 %, FlopsRate: 0.002 %
                                       BinaryOp111 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.007 %
                                       BinaryOp113 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.007 %
                                       BinaryOp136 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.003 %
                                       BinaryOp161 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.004 %
                                       BinaryOp163 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.004 %
                                       BinaryOp212 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.015 %
                               Softmax264_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.000 %
                                   conv2d_55.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.001 %
                                   conv2d_56.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.001 %
                                   conv2d_57.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.370 %
                                   conv2d_79.tmp_0 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.046 %
                          depthwise_conv2d_0.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.416 %
                               hardsigmoid_3.tmp_0 	[ReLU6] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.007 %
                               hardsigmoid_7.tmp_0 	[ReLU6] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.015 %
                                hardswish_10.tmp_0 	[UnaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.088 %
                                        p2o.Add.25 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.029 %
                                        p2o.Add.35 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.029 %
                                        p2o.Add.45 	[BinaryOp] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.029 %
                      p2o.Add.87__matmul_converted 	[Convolution] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.032 %
                               p2o.Mul.15_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.162 %
                               p2o.Mul.39_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.096 %
                               p2o.Mul.45_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.184 %
                               p2o.Mul.51_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.184 %
                                p2o.Mul.9_raster_0 	[Raster] run 1 average cost 0.002000 ms, 0.259 %, FlopsRate: 0.059 %
                                       BinaryOp138 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.003 %
                                       BinaryOp186 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.008 %
                                       BinaryOp210 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                       BinaryOp235 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                       BinaryOp237 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                        BinaryOp62 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.002 %
                                        BinaryOp64 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.002 %
                                        BinaryOp86 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.007 %
                                        BinaryOp88 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.007 %
                                   conv2d_59.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.555 %
                                   conv2d_63.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.021 %
                                   conv2d_68.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.155 %
                                   conv2d_72.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.155 %
                                   conv2d_75.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.032 %
                                   conv2d_76.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.032 %
                                   conv2d_80.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.046 %
                                   conv2d_83.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.217 %
                                   conv2d_88.tmp_0 	[Convolution] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.802 %
                               hardsigmoid_0.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.001 %
                               hardsigmoid_2.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.007 %
                               hardsigmoid_4.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.003 %
                               hardsigmoid_5.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.004 %
                               hardsigmoid_6.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.008 %
                               hardsigmoid_8.tmp_0 	[ReLU6] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.015 %
                                 hardswish_1.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.088 %
                                hardswish_12.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.096 %
                                hardswish_13.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_14.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_15.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_16.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                hardswish_17.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                 hardswish_2.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.059 %
                                 hardswish_8.tmp_0 	[UnaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.073 %
                                        p2o.Add.55 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.029 %
                                         p2o.Add.7 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.022 %
                                        p2o.Add.83 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.029 %
                 p2o.GlobalAveragePool.11_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                 p2o.GlobalAveragePool.13_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.3_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.9_raster_0 	[Pooling] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.000 %
                                        p2o.Mul.15 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.162 %
                               p2o.Mul.21_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.162 %
                                        p2o.Mul.27 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.073 %
                               p2o.Mul.27_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.073 %
                                         p2o.Mul.3 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.044 %
                               p2o.Mul.33_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.088 %
                                p2o.Mul.3_raster_0 	[Raster] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.044 %
                                        p2o.Mul.45 	[BinaryOp] run 1 average cost 0.003000 ms, 0.388 %, FlopsRate: 0.184 %
                                   conv2d_54.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.740 %
                                   conv2d_61.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.740 %
                                   conv2d_62.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.740 %
                                   conv2d_64.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.021 %
                                   conv2d_65.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.986 %
                                   conv2d_67.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.155 %
                                   conv2d_71.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.155 %
                                   conv2d_77.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 1.233 %
                                   conv2d_78.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 1.479 %
                                   conv2d_84.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.217 %
                                   conv2d_87.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.802 %
                                   conv2d_91.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.802 %
                                   conv2d_92.tmp_0 	[Convolution] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.802 %
                          depthwise_conv2d_1.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.624 %
                                 hardswish_4.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.162 %
                                 hardswish_6.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.162 %
                                 hardswish_7.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.073 %
                                 hardswish_9.tmp_0 	[UnaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.088 %
                 p2o.GlobalAveragePool.19_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.1_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.5_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                  p2o.GlobalAveragePool.7_raster_0 	[Pooling] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.000 %
                                        p2o.Mul.21 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.162 %
                                        p2o.Mul.33 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.088 %
                                        p2o.Mul.39 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.096 %
                                        p2o.Mul.51 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.184 %
                                         p2o.Mul.9 	[BinaryOp] run 1 average cost 0.004000 ms, 0.517 %, FlopsRate: 0.059 %
                                   conv2d_58.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 1.109 %
                                   conv2d_60.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.740 %
                                   conv2d_81.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 1.479 %
                                   conv2d_82.tmp_0 	[Convolution] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 3.205 %
                          depthwise_conv2d_2.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.832 %
                          depthwise_conv2d_3.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 1.541 %
                          depthwise_conv2d_7.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 2.311 %
                          depthwise_conv2d_8.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 2.504 %
                                 hardswish_0.tmp_0 	[UnaryOp] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.088 %
                                        p2o.Add.73 	[BinaryOp] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.029 %
             p2o.Add.87__matmul_converted_raster_0 	[Raster] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.015 %
                 p2o.GlobalAveragePool.15_raster_0 	[Pooling] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.000 %
                 p2o.GlobalAveragePool.17_raster_0 	[Pooling] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.000 %
                                     p2o.MaxPool.1 	[Pooling] run 1 average cost 0.005000 ms, 0.647 %, FlopsRate: 0.367 %
                                       BinaryOp188 	[BinaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.008 %
                                   conv2d_66.tmp_0 	[Convolution] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 2.712 %
                                   conv2d_70.tmp_0 	[Convolution] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 2.712 %
                                   conv2d_73.tmp_0 	[Convolution] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 2.712 %
                          depthwise_conv2d_5.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 4.237 %
                          depthwise_conv2d_6.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 1.926 %
                                hardswish_11.tmp_0 	[UnaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.191 %
                                 hardswish_3.tmp_0 	[UnaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.162 %
                                 hardswish_5.tmp_0 	[UnaryOp] run 1 average cost 0.006000 ms, 0.776 %, FlopsRate: 0.162 %
                                   conv2d_69.tmp_0 	[Convolution] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 2.712 %
                                   conv2d_74.tmp_0 	[Convolution] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 1.233 %
                                   conv2d_90.tmp_0 	[Convolution] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 6.163 %
                          depthwise_conv2d_4.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.007000 ms, 0.906 %, FlopsRate: 4.237 %
                                   conv2d_86.tmp_0 	[Convolution] run 1 average cost 0.008000 ms, 1.035 %, FlopsRate: 6.163 %
                                   conv2d_94.tmp_0 	[Convolution] run 1 average cost 0.008000 ms, 1.035 %, FlopsRate: 6.163 %
                          depthwise_conv2d_9.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.008000 ms, 1.035 %, FlopsRate: 4.815 %
                         depthwise_conv2d_10.tmp_0 	[ConvolutionDepthwise] run 1 average cost 0.009000 ms, 1.164 %, FlopsRate: 4.815 %
                                   conv2d_85.tmp_0 	[Convolution] run 1 average cost 0.011000 ms, 1.423 %, FlopsRate: 3.205 %
                                   conv2d_93.tmp_0 	[Convolution] run 1 average cost 0.018000 ms, 2.329 %, FlopsRate: 6.163 %
                                   conv2d_89.tmp_0 	[Convolution] run 1 average cost 0.019000 ms, 2.458 %, FlopsRate: 6.163 %
                                   conv2d_53.tmp_0 	[Convolution] run 1 average cost 0.030000 ms, 3.881 %, FlopsRate: 2.496 %
Avg= 0.773000 ms, OpSum = 0.567000 ms min= 0.773000 ms, max= 0.773000 ms

 
./quantized.out cls.mnn cls-quant.mnn imageInputConfig.json
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1465: >>> modelFile: /home/vca1/swq/code/PaddleOCR/models/zwd_mnn_fp32_new/cls.mnn
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1466: >>> preTreatConfig: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/imageInputConfig.json
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1467: >>> dstFile: /home/vca1/swq/code/PaddleOCR/models/zwd_quantout/cls-quant.mnn
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1495: Calibrate the feature and quantize model...
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:306: Use feature quantization method: KL
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:307: Use weight quantization method: MAX_ABS
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:327: feature_clamp_value: 127
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:328: weight_clamp_value: 127
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:332: winogradOpt only be available under EMA
CPU Group: [ 29  55  27  65  5  37  75  47  19  57  17  67  7  39  10  77  49  20  59  30  71  21  31  41  13  51  23  61  1  33  69  43  15  53  25  63  3  35  73  45  28  54  26  64  4  36  74  46  18  56  16  66  6  38  76  48  58  68  8  11  70  9  40  12  79  50  22  60  0  32  78  42  14  52  24  62  2  34  72  44 ], 800000 - 3900000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/Helper.cpp:143: used dataset num: 1
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:830: fake quant weights done.
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
CollectFeatureDistribution: 100.00 %
Invalid Tensor, the session may not be ready
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
Compute Shape Error for conv2d_53.tmp_0
code=3 in onForward, 564 
computeDistance: 100.00 %

Debug info:

Can't find input extraTensorDescribe for hardswish_0.tmp_0
Can't find input extraTensorDescribe for hardswish_1.tmp_0
Can't find input extraTensorDescribe for hardswish_2.tmp_0
Can't find input extraTensorDescribe for hardswish_3.tmp_0
Can't find input extraTensorDescribe for hardswish_4.tmp_0
Can't find input extraTensorDescribe for hardswish_5.tmp_0
Can't find input extraTensorDescribe for hardswish_6.tmp_0
Can't find input extraTensorDescribe for hardswish_7.tmp_0
Can't find input extraTensorDescribe for hardswish_8.tmp_0
Can't find input extraTensorDescribe for hardswish_9.tmp_0
Can't find input extraTensorDescribe for hardswish_10.tmp_0
Can't find input extraTensorDescribe for hardswish_11.tmp_0
Can't find input extraTensorDescribe for hardswish_12.tmp_0
Can't find input extraTensorDescribe for hardswish_13.tmp_0
Can't find input extraTensorDescribe for hardswish_14.tmp_0
Can't find input extraTensorDescribe for hardswish_15.tmp_0
Can't find input extraTensorDescribe for hardswish_16.tmp_0
Can't find input extraTensorDescribe for hardswish_17.tmp_0
[17:59:42] /home/vca1/zwd/MNN-3.0.0/tools/quantization/calibration.cpp:1503: Quantize model done!

从这个 unary 不支持的 log 看上去，本地代码没有更新并重编。

先更新一下试试吧

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
quantization
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@jxt1234 @Tmn07