Hi everyone!
I am currently working on a project focusing on RNA modifications. I built and optimized a custom 1D-CNN ensemble architecture in PyTorch.
However, in the lab we hit a technical roadblock while trying to deploy our trained weights into Dorado via ONNX for real-time basecalling. During the compilation/serialization stage, Dorado throws this error:
It looks like the Dorado framework expects a very specific model signature signature/dimension structure [28] and requires an official digital signature validation core (ON_SVC) to parse custom graphs.
Has anyone here successfully integrated a custom neural network model directly into Dorado? Is there a known protocol, workaround, or a specific script to sign and format the weights so the Dorado compiler accepts them?
We would highly appreciate any tips, workarounds, or documentation from anyone who has managed to bypass this restriction!
Thank you so much!
I am currently working on a project focusing on RNA modifications. I built and optimized a custom 1D-CNN ensemble architecture in PyTorch.
However, in the lab we hit a technical roadblock while trying to deploy our trained weights into Dorado via ONNX for real-time basecalling. During the compilation/serialization stage, Dorado throws this error:
It looks like the Dorado framework expects a very specific model signature signature/dimension structure [28] and requires an official digital signature validation core (ON_SVC) to parse custom graphs.
Has anyone here successfully integrated a custom neural network model directly into Dorado? Is there a known protocol, workaround, or a specific script to sign and format the weights so the Dorado compiler accepts them?
We would highly appreciate any tips, workarounds, or documentation from anyone who has managed to bypass this restriction!
Thank you so much!