TFLite converter effecientnet-b0 error #511
Comments
|
@n-berezina-nn Наталья, не могли бы вы помочь разобраться почему модель efficientnet-b0 из OpenModelZoo не конвертируется в формат tflite? |
|
@n-berezina-nn, @FenixFly, сконвертировала модель с использованием следующего скрипта: import tensorflow as tf
converter = tf.lite.TFLiteConverter.from_saved_model('./efficientnet-b0/saved_model/')
converter.target_spec.supported_ops = [
tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS
]
converter.allow_custom_ops=True
tflite_model = converter.convert()
with open('model.tflite', 'wb') as f:
f.write(tflite_model)Запускаю вывод средствами вот такой командной строки: python3 ./inference_tensorflowlite.py -m ./efficientnet-b0/model.tflite -i ./data/ -b 1 -t classification --output_names logits -l ./labels/image_net_synset.txt --input_names sub[1,224,224,3]Получаю следующий вывод: 2024-03-20 22:07:52.309100: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
2024-03-20 22:07:52.309335: I external/local_tsl/tsl/cuda/cudart_stub.cc:32] Could not find cuda drivers on your machine, GPU will not be used.
2024-03-20 22:07:52.311233: I external/local_tsl/tsl/cuda/cudart_stub.cc:32] Could not find cuda drivers on your machine, GPU will not be used.
2024-03-20 22:07:52.334658: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-03-20 22:07:52.738092: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
[ INFO ] Loading network files:
/home/itmm/Documents/kustikova_v/public/efficientnet-b0/efficientnet-b0/model.tflite
INFO: Created TensorFlow Lite delegate for select TF ops.
INFO: TfLiteFlexDelegate delegate: 64 nodes delegated out of 270 nodes with 64 partitions.
2024-03-20 22:07:52.998300: E tensorflow/core/framework/node_def_util.cc:676] NodeDef mentions attribute use_inter_op_parallelism which is not in the op definition: Op<name=DepthwiseConv2dNative; signature=input:T, filter:T -> output:T; attr=T:type,allowed=[DT_HALF, DT_BFLOAT16, DT_FLOAT, DT_DOUBLE]; attr=strides:list(int); attr=padding:string,allowed=["SAME", "VALID", "EXPLICIT"]; attr=explicit_paddings:list(int),default=[]; attr=data_format:string,default="NHWC",allowed=["NHWC", "NCHW"]; attr=dilations:list(int),default=[1, 1, 1, 1]> This may be expected if your graph generating binary is newer than this binary. Unknown attributes will be ignored. NodeDef: {{node DepthwiseConv2dNative}}
[ INFO ] Shape for input layer sub:0: 1x224x224x3
[ INFO ] Preparing input data: ['/home/itmm/Documents/kustikova_v/data/']
[ INFO ] Starting inference (1 iterations)
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
[ ERROR ] Traceback (most recent call last):
File "/home/itmm/Documents/kustikova_v/upstream/dl-benchmark/src/inference/./inference_tensorflowlite.py", line 286, in main
result, inference_time = inference_tflite(interpreter, args.number_iter, io.get_slice_input, args.time)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/itmm/Documents/kustikova_v/upstream/dl-benchmark/src/inference/./inference_tensorflowlite.py", line 191, in inference_tflite
interpreter.allocate_tensors()
File "/home/itmm/miniconda/envs/tflite_converter_env/lib/python3.11/site-packages/tensorflow/lite/python/interpreter.py", line 531, in allocate_tensors
return self._interpreter.AllocateTensors()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Encountered unresolved custom op: swish_f320.
See instructions: https://www.tensorflow.org/lite/guide/ops_custom Node number 2 (swish_f320) failed to prepare.Encountered unresolved custom op: swish_f320.
See instructions: https://www.tensorflow.org/lite/guide/ops_custom Node number 2 (swish_f320) failed to prepare.Если все правильно понимаю, то это означает, что конвертер посчитал, что оператор |
|
@FenixFly, @n-berezina-nn, сравнила через netron модель из OMZ в формате saved_model, которую мы хотим сконвертировать, и присланную модель в формате tflite. В модели OMZ есть явный слой swish_f32 (схема ниже), в присланной модели таких преобразований вообще нет. Это означает, что либо для конвертации использована другая модель, либо эту модель как-то преобразовывали перед тем, как сконвертировать.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Пытаюсь конвертировать TF модель efficientnet-b0 из OpenModelZoo.
tflite_converter.pyвыдает ошибку с неподдерживаемым слоем__inference_swish_f32_730.Пробовал с разными tensorflow, последний раз с tensorflow==2.12.0 и tensorflow-addons==0.19 tensorflow-estimator==2.12.0 tensorflow-probability==0.19
Строка конвертации:
Вывод скрипта с ошибкой (часть, полный вывод по ссылке: https://gist.github.com/FenixFly/3c523abc679934a7df6b67ad006c7ad6 ):
The text was updated successfully, but these errors were encountered: