Dilshod Tadjibaev

OK. Refactoring is done and all tests are passing. Here is an example of ONNX-IR with typed config: ``` OnnxGraph { nodes: [ Conv2d { name: "conv2d1", inputs: [ Argument...

CCing: @laggui, @skewballfox, tiruka, @hexd0t, @nathanielsimard

@skewballfox I should have made requirements more clear. Currently there are two requirements: 1. **Burn's Shape Handling**: For Burn, we aim to avoid static shapes entirely (Burn's own graph will...

> For 1, could you point me to some relevant examples of what the syntax might look like? or just code out some examples here? I'm mainly trying to get...

> Even if the shapes can be changed at runtime, it still seems reasonable to want to initialize the tensors with some expected capacity. This feature can be implemented if...

Update: I have lots of work in this area that is in my pipeline, sitting on my desktop. Hopefully this week or next I'll be able to submit a PR....

OK. Refactoring is done and all tests are passing. Here is an example of ONNX-IR with typed config: ``` OnnxGraph { nodes: [ Conv2d { name: "conv2d1", inputs: [ Argument...

Slight change of plans. We will have structure inside Node enums. **Current Structure** ```rust // In src/ir/node.rs - Node enum with struct variants pub enum Node { ArgMax { name:...

Quick update I am able to convert `yolo11x_opset16.onnx` using this WIP PR: #3381

CLIP ViT-B-32 is buildable with https://github.com/tracel-ai/burn/pull/3560 (still under review)