Name | Modified | Size | Downloads / Week |
---|---|---|---|
Parent folder | |||
README.md | 2023-06-13 | 1.0 kB | |
v1.5.1 source code.tar.gz | 2023-06-13 | 45.4 MB | |
v1.5.1 source code.zip | 2023-06-13 | 48.3 MB | |
Totals: 3 Items | 93.6 MB | 0 |
Added
- Support Python API
- Support AVX-VNNI and ARMv9 instruction set
- Support Intel Desktop GPU (float16 and float32)
- Support Windows on arm platform
- Support more operators : Random, Sin, Cos, Einsum, Elu, UnPooling, Flatten, ConvertColor, BilateralSliceApply, Lut
- Support more networks : ViTAE, CMT, EfficientFormer, ConvTT, Wenet, NFM, AFM, ONN, wide&deep, DeepFM, MMOE, etc
- Improve multi-threads parallel inference performance on CPU
- Add simple chinese deployment guide
- Support model file compatibility
- Support using outer memory(CPU array or OpenCL cl_mem) by using SetInputOutput API
- Support data type and format transform by using C API
Changed
- TensorDesc's dim array is changed to 20.
- Remove FILE macro usage and warning log under release mode
- change enum data and operator parameter size
Fixed
- Fix GPU resize bug
- Fix GPU concurrent inference bug
- Fix ONNX converter bug
- Add missed chinese automatic speech recognition model