2.1.1

[2.1.1] - 2024-11-07

Fixed

• Ordering of inputs now matches ONNX model consistently
• Loading of models with constants over 500mb
• Serialization of models over 2GB in size
• Issues in example code in documentation
• Shape inference for Split and Select operators
• Profiler tags for operators are now consistent with lower overhead
• CopyOutput can now be used on GPUPixel backend if the tensor is the correct shape