Hi,
I'm working with the i.MX 8M Plus EVK and VX delegate. I tried to run a full-int quantized tflite model using tflite_runtime version 2.6
delegate = [tflite.load_delegate('/usr/lib/libvx_delegate.so')]
interpreter = tflite.Interpreter(model_path=path, experimental_delegates=delegate)
After allocating and setting input tensor, interpreter.invoke() generates following output:
Vx delegate: allowed_builtin_code set to 0. Vx delegate: error_during_init set to 0. Vx delegate: error_during_prepare set to 0. Vx delegate: error_during_invoke set to 0.
W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. W [HandleLayoutInfer:257]Op 18: default layout inference pass. E [vsi_nn_OpOptimize:253]Optimize node RESHAPE fail E [optimize_node_backward:260]Backward optimize node[63] RESHAPE fail ERROR: Failed to verify graph
Following OpCodes are used in the already converted tflite model:
(Conversion is also done with tensorflow 2.6.0 lite converter)
index | builtin_code | custom_code | version |
---|---|---|---|
0 | PAD | 2 | |
1 | RESHAPE | 1 | |
2 | CONV_2D | 3 | |
3 | MAX_POOL_2D | 2 | |
4 | TRANSPOSE_CONV | 3 | |
5 | ADD | 2 |
As attached in the i.mx machine learning user guide, the listed operations above are all supported by the ovxlib.
What could cause this error? Can someone help?
Thanks and best regards
Hello,
I have the same issue with both Thunder and Lightning MoveNet models (https://tfhub.dev/google/movenet/singlepose/thunder/4) (https://tfhub.dev/google/movenet/singlepose/lightning/4). I attach a .txt file with logs.