Tags: we0091234/DALI
Tags
Fix fp16 bug from NVIDIA#1129 and add fp16 test case (NVIDIA#1160) Initializing member variable `OutputType padding_value = 0;` for `__half` in the host code produced wrong results. `padding_val` is now a fp32 that is casted to `OuputType` in the CUDA kernel Signed-off-by: Serge Panev <spanev@nvidia.com>
Change version to 0.14.0dev Signed-off-by: Krzysztof Lecki <klecki@nvidia.com>
Change version to 0.13.0 Signed-off-by: Krzysztof Lecki <klecki@nvidia.com>
Reduced batch size to avoid out of memory condition in 19.07 containe… …r. (NVIDIA#1089) Signed-off-by: Michal Zientkiewicz <michalz@nvidia.com>
Reduced batch size to avoid out of memory condition in 19.07 containe… …r. (NVIDIA#1089)
Change version to 0.13.0dev Signed-off-by: Krzysztof Lecki <klecki@nvidia.com>
Change version to 0.12.0 Signed-off-by: Krzysztof Lecki <klecki@nvidia.com>
Add as_tensor with provided shape method to python API (NVIDIA#953) - user can cast TensorList to Tensor with provided shape from python API, as long as TensorList is continuous and volumes matches Signed-off-by: Janusz Lisiecki <jlisiecki@nvidia.com>
Add as_tensor with provided shape method to python API (NVIDIA#953)
PreviousNext