search

IBM / fhe-toolkit-linux

IBM Fully Homomorphic Encryption Toolkit For Linux. This toolkit is a Linux based Docker container that demonstrates computing on encrypted data without decrypting it! The toolkit ships with two demos including a fully encrypted Machine Learning inference with a Neural Network and a Privacy-Preserving key-value search.
MIT License
1.43k stars 158 forks source link

Error : nnp.init_from_onnx #193

Closed Garmin-Qian closed 1 year ago

Garmin-Qian commented 1 year ago

Hello! I saved my model. But when I use the following command, the program reports an error.

111

Garmin-Qian commented 1 year ago

The framework of the model has no errors because it can be trained successfully.

Garmin-Qian commented 1 year ago

When I delete "bias=False", I solve this problem. But I meet the new problem.

"RuntimeError: Bad data type 7 for tensor"

boland25 commented 1 year ago

I think this has to be an area where @dubek or @aehud get involved.

aehud commented 1 year ago

Hi. Thanks for notifying these issues.

Regarding the convolution, you are right, we currently don't support convolution without bias. We plan to add this support soon. Also we plan to improve error messages, and in fact the latest version, 1.5.0.3, already has some improved messages. You are welcome to check it.

Regarding the second issue: we currently support floating point datatype for the values provided in the onnx file, e.g., for weights of the various layers, and also as part of the activation layers definition. The '7' indicates int64 types. Here too we plan to extend this support and improve error messages.

In the meantime:

Thanks.