Releases: ENOT-AutoDL/onnx2torch
Releases · ENOT-AutoDL/onnx2torch
v1.5.4
v1.5.3
v1.5.2
v1.5.1
v1.5.0
You've been waiting, you've asked... And here it is a new release(1.5.0)!
In this release:
- Fixed shape inference of big models (now you can load your lovely big language model, like gptj6B);
- Fixed
BarchNorm
converter accordingonnx
specs; - Added asymmetric padding for both
Conv
andMaxPool
.
Thank you for your discussions and issues! We always ready to help you!
v1.4.1
v1.4.0
🚀 New day - new release! 🚀
New features in release 1.4.0:
-
Added operations:
- Einsum,
- Reciprocal,
- Neg,
- Prelu,
- Mean,
- Min,
- Max,
- CumSum,
- Sum.
-
Fixes:
- GlobalAveragePool compatible with PyTorch FX QAT,
- Compatibility with torch 1.12,
- Allow to ignore bs+ch dimensions for size input in Resize.
- Check roi only for tf_crop_and_resize mode.
-
Code style and workflows:
- Pre-commit hook,
- Black code-style now is used in project,
- Linter and formatter with precommit hook,
- Code style checks for PRs in Github actions.
We thank all contributors for their work! Together we will make onnx2torch great! ⭐ 10^6
v1.3.0
v1.2.5
v1.2.0
New features in release 1.2.0:
- Support
VIT
,SWIN
,FasterRcnn
models; - add
RoiAlign
operation; - Add
Floor
,Ceil
,Round
and trigonometric functions; - New model tests for classification, detection and segmentation;
Fixes:
- Fix
ScatterND
and its speed; - Fix names of plaseholders;
- Fix optional arguments if last arguments are not passed;
- Fix checking of empty inputs;
- Fix avg pool operation;