Hello Nx community!

I’m here as a noob in the field of AI (and python), trying to accomplish a task with my favourite language. I’d like to use an exported ONNX model (exported from a torch model, built in python, which I can load from the filesystem). Someone before me has written some python code to execute the torch model, which I would like to do in elixir. The gist boils down to the following:

import pandas as pd
import numpy as np
import torch

model = torch.load('/path/to/model.torch', map_location=torch.device('cpu'))
pred = {}

for index, data_frame in some_list_of_data_frames:
  feats = torch.from_numpy(np.array(data_frame, dtype=np.float32))
  with torch.no_grad():
    pred[index] = model.forward(feats.unsqueeze(0)).detach().cpu().numpy()[0]

result = pd.DataFrame.from_dict(pred, orient="index", columns=["some", "set", "of", "columns"])

The data frames are built somehow, but that’s for later (with explorer, I suppose). I’d like to know what a comparable elixir implementation would be, if at all possible. I manage to load the ONNX model with the axon_onnx lib, which spits out the axon model and a map of params, which is a start. But then I don’t really know how to mimic the torch.from_numpy(), torch.no_grad(), model.forward(...).detach().cpu().numpy() and pd.DataFrame.from_dict(...). If anyone has tips to share, I’d be very grateful