Dilshod Tadjibaev

@LaurentMazare has confirmed Candle supports 0D tensors: > Zermelo Fraenkel: Scalars values (tensors with 0 dimension) should be supported. Empty tensors (multiple dimensions but with one of them being zero)...

cc @ashdtu

@laggui found that Ndarray supports 0dim arrays: https://docs.rs/ndarray/latest/ndarray/type.Array0.html

@nathanielsimard and I had an offline conversation. Here's a summary of the conversation for others: We discussed the need to support scalar tensors in the Burn deep learning framework. While...

This project should be related: https://github.com/justinjohn0306/so-vits-svc-4.0-v2

bool_full implementation: ```rust fn bool_full( shape: Shape, value: bool, device: &Device, ) -> BoolTensor { B::int_equal_elem( B::int_zeros(shape, device), if value { 0.elem() } else { 1.elem() }, ) } ```

We need full working for bool tensor because ConstantOfShape can initialize with bool values and we can't use full because of this error: ``` Compiling onnx-tests v0.14.0 (/Users/dilshod/Projects/burn/crates/burn-import/onnx-tests) error[E0277]: the...

Workaround to get full for bool tensors: ```rust // All true Tensor::::ones(shape, &device).bool(); // All false Tensor::::zeros(shape, &device).bool(); ```

@laggui , you mentioned you were working in this area. Do you think we are close making it work?

Thanks, @AdrianEddy, for filing this issue. This helps us prioritize the ONNX ops. OK. I have checked the ONNX file and it contains lots of nodes (> 10K) but with...