v1.14.0

ort 1.14 is updated for ONNX Runtime v1.14.0. Please update to 1.14 as soon as possible, as this release contains important fixes and behaviour changes.

Changes (compared to v1.13)

• Simplified the process for downloading models.
• Made environments easier to use.
  • You can now use Environment::default() to return the global environment, or create a dummy one if one does not exist.
  • Environment::into_arc() automatically converts an Environment into an Arc<Environment> for cleaner code.
• Fixed an issue where tensors being passed to ONNX Runtime were not laid out contiguously in memory before being passed to the session, and using methods like Array::slice on an array before passing it to ort would result in unexpected results.
• Reduced bloat by making the download strategy optional, but enabled by default. See the README for more info. Developers of libraries using ort should consider using default-features = false.
• Added support for zero-copy deserialization for sessions loaded from memory via SessionBuilder::with_model_from_memory. This feature only works with .ort models.
  • Note that SessionBuilder::with_model_from_memory<'s>(&'s array) returns an InMemorySession<'s>, not a Session.
• Now using AllocatorType::Device instead of AllocatorType::Arena to potentially improve memory usage.
• Added support for the ROCm & DirectML execution providers (still working out ROCm build issues, see #16)
• Fixed some issues occurring with the compile strategy.
• Added memory pattern optimization with Session::with_memory_pattern
• ExecutionProvider::x_if_available has been removed, as the functionality is identical to ExecutionProvider::x.
• Removed the disable-build-script feature. Libraries on docs.rs using ort will no longer need to enable the disable-build-script feature in docs.rs build; ort will automatically detect if it is running in docs.rs and disable the build script.
• Removed the prefer-* features, as they were not very useful and potentially added unneeded complexity.
• Added support for scalar inputs (0-dimensional arrays) for models like Silero VAD.