TensorRT Conversion Error

[Sayyam-Jain]

I'm trying to convert the provided pth (small-satrn) model to tensorrt. But it shows the following error.

2020-09-14 14:05:55,032 - INFO - Use GPU 0
2020-09-14 14:05:55,032 - INFO - Set cudnn deterministic False
2020-09-14 14:05:55,032 - INFO - Set cudnn benchmark True
2020-09-14 14:05:55,032 - INFO - Set seed 1111
2020-09-14 14:05:55,032 - INFO - Build model
2020-09-14 14:05:55,036 - INFO - GResNet init weights
2020-09-14 14:05:55,348 - INFO - TransformerEncoder init weights
2020-09-14 14:05:55,891 - INFO - TransformerDecoder init weights
2020-09-14 14:05:56,001 - INFO - TransformerHead init weights
2020-09-14 14:05:57,158 - INFO - Load checkpoint from their.pth
2020-09-14 14:05:57,292 - INFO - Convert to trt engine with fp32
Dummy Input <class 'list'>
Traceback (most recent call last):
File "tools/deploy/export.py", line 99, in
main()
File "tools/deploy/export.py", line 92, in main
int8_mode=int8_mode, int8_calibrator=int8_calibrator)
File "/home/sayyam/anaconda3/envs/virtualenv/lib/python3.7/site-packages/volksdep-3.0.0-py3.7.egg/volksdep/converters/torch2trt.py", line 40, in torch2trt
torch2onnx(model, dummy_input, f)
File "/home/sayyam/anaconda3/envs/virtualenv/lib/python3.7/site-packages/volksdep-3.0.0-py3.7.egg/volksdep/converters/torch2onnx.py", line 46, in torch2onnx
output = model(dummy_input)
File "/home/sayyam/anaconda3/envs/virtualenv/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "tools/deploy/../../vedastr/models/model.py", line 22, in forward
out = self.head(x, inputs[1])
File "/home/sayyam/anaconda3/envs/virtualenv/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "tools/deploy/../../vedastr/models/heads/transformer_head.py", line 75, in forward
tgt = self.text_embedding(texts)
File "tools/deploy/../../vedastr/models/heads/transformer_head.py", line 54, in text_embedding
tgt = self.embedding(texts)
File "/home/sayyam/anaconda3/envs/virtualenv/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/home/sayyam/anaconda3/envs/virtualenv/lib/python3.7/site-packages/torch/nn/modules/sparse.py", line 114, in forward
self.norm_type, self.scale_grad_by_freq, self.sparse)
File "/home/sayyam/anaconda3/envs/virtualenv/lib/python3.7/site-packages/torch/nn/functional.py", line 1724, in embedding
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
TypeError: embedding(): argument 'indices' (position 2) must be Tensor, not list

Similar error (list has no attribute size) occurs when using TPS_Resnet-BiLSTM_Attention Model.
Please help.
Thanks

[ChaseMonsterAway]

We just test resnet ctc model. There are some ops are not supported by TensorRT.

[ChaseMonsterAway]

We will check it in the future. Please keep an eye on our work, volksdep.

[chophilip21]

any update on this issue?