Releases: pykeio/ort
Releases · pykeio/ort
v1.14.5
v1.14.4
Changes
- 5bb47ef Fix execution providers not being registered on
load-dynamic
(#34) - e147325 Attempts to register all execution providers specified instead of returning after the first successful registration, matching the behaviour of the ONNX Runtime Python API.
- 1127dad Fix the CUDA execution provider failing to compile on Linux ARM64.
- 0e7faaa Fix the TensorRT execution provider returning as "not available" when it is really available.
- 7c0103b Allow for a flat library directory structure with
ORT_STRATEGY=system
(not requiring aninclude/
andlib/
dir) - 6e9dce0 Fix the
assert_non_null_pointer
function producing a misleadingPointerShouldBeNull
error. - 41b6ea0 Increase the timeout of the binary download in case of slow connections. (#33)
v1.14.3
v1.14.2
This release updates to ONNX Runtime v1.14.1, containing fixes for performance regressions with transformer models. Additionally, issues with static linking on Windows and building on macOS have been fixed, and a new feature, load-dynamic
, eliminates the shared library hell via dlopen()
with an env-controllable path instead of dynamic linking.
Changes
- 566ef7f Update to ONNX Runtime v1.14.1
- 930edb6 Adds the
load-dynamic
feature, eliminating the shared library hell - c7b668e Fixes incorrect incompatible provider errors on macOS (thanks @ williamhogman for pointing this out)
- 8889616 Replaces
OrtSessionOptionsAppendExecutionProvider_ROCm
withOrtApi().SessionOptionsAppendExecutionProvider_ROCM
to fix a linking issue. - 1afd392 f6135fc Allow for compiling on Windows x86.
- 7f8c551 Fix
copy-dylibs
failing with static linking.
Static linking fixes
Static linking is confirmed working with microsoft/onnxruntime#v1.14.1 on Windows x64.
v1.14.1
- Libraries using
default-features = false
will no longer encounter build errors when runningcargo build
orcargo check
.- Instead, a linking error will occur only when building a binary, informing the user to enable the
download-binaries
feature or use thesystem
strategy.
- Instead, a linking error will occur only when building a binary, informing the user to enable the
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 anEnvironment
into anArc<Environment>
for cleaner code.
- You can now use
- 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 usingdefault-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 anInMemorySession<'s>
, not aSession
.
- Note that
- Now using
AllocatorType::Device
instead ofAllocatorType::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 toExecutionProvider::x
.- Removed the
disable-build-script
feature. Libraries on docs.rs using ort will no longer need to enable thedisable-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.
v1.14.0-beta.0
- Updated for ONNX Runtime v1.14.0
- Now using
AllocatorType::Device
instead ofAllocatorType::Arena
to potentially improve memory usage. - Support for the ROCm execution provider
- 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 toort
would result in unexpected results.
v1.14.0-alpha.0
Based on ONNX Runtime v1.13.1.
This release includes the following changes:
- Support for the DirectML execution provider
- The Profiling API is now only available with the
profiling
feature. Session::with_memory_pattern()
to enable/disable memory pattern optimization.- Model downloading has been simplified. Now e.g.
GPT2::GPT2LmHead
can be downloaded directly. - Added
Environment::default()
, which will return a threads existing environment, or create one if it does not exist. - Added
Environment::into_arc()
to convert theEnvironment
into anArc<Environment>
. disable-build-script
has been removed. Crates usingort
no longer need it to get docs.rs builds working.- The
prefer-compile-strategy
,prefer-system-strategy
, andprefer-dynamic-libs
features have been removed. - Fixed scalar inputs for models like Silero VAD.