Dilshod Tadjibaev

Yeah it'd be great if we have example. Meanwhile you can see some basic tests: https://github.com/tracel-ai/burn/blob/main/crates/burn-dataset/src/dataset/dataframe.rs

BTW, the API is available via a `dataframe` feature flag.

@skewballfox, do you think this is related to #1857?

> Do you guys off the top of your head know burn-graph side what is supposed to happen to input args with no parents? are they inlined, stored as variables?...

> if I convert the initializer to a constant and disable constant lifting, I run into an error where an `Int` constant tensor is added to the imported model, however,...

We should probably fix this to support older opset. As a work around, can you convert to a new opset? There are some tools to upgrade the onnx file.

This part needs to be fixed: We need to support Gather, Unsqueeze, and Concat operations on Shape

Used `onnx-simplifier` (https://github.com/daquexian/onnx-simplifier) which removes lots of these crazy gather reshape operation, which were due to lacking capabilities of the older ONNX OpSet versions. Here is updated version: [face_detector_opt.onnx.zip](https://github.com/user-attachments/files/16118060/face_detector_opt.onnx.zip) Output...

Submitted a PR for the slice issue: https://github.com/tracel-ai/burn/pull/1989 But I stumbled another bug: https://github.com/tracel-ai/burn/issues/1983 which I am working to fix. Overall, I think I am very close to fixing it....

I got the face-onnx building, however, there is a run-time error that's probably introduced recently when we were refactoring tensors. @laggui did some work in the Tensor random initialization. Maybe...