tt-metal: YoloV4 ttnn.conv2d fails with "Input tensor must be width sharded" error
Describe the bug On implementing the concat op using sharded data as input as in 10808 12182, facing “Input tensor must be width sharded” error in the following conv.
To Reproduce Steps to reproduce the behavior:
- Checkout keerthanar/yolov4_shard_conv_issue
- Run
pytest tt-metal/tests/ttnn/unit_tests/operations/test_concat_conv.py::test_concat_conv_yolov4 - The following error will occur,
RuntimeError: TT_FATAL @ ../ttnn/cpp/ttnn/operations/data_movement/tilize_with_val_padding/device/tilize_with_val_padding_op.cpp:37: input_tensor_a.memory_config().memory_layout == TensorMemoryLayout::WIDTH_SHARDED E info: E Input tensor must be width sharded
Please complete the following environment information:
- WH n150
About this issue
- Original URL
- State: closed
- Created 2 years ago
- Comments: 22 (22 by maintainers)
Commits related to this issue
- #12387: Yolov4 sharded concat conv issue — committed to tenstorrent/tt-metal by keerthana-r-mcw 2 years ago
- #12387: Workaround to_layout for height sharded tensor Conv2d will potentially trigger ttnn::to_layout for the input tensor in order to tilize the input. In cases where the input needs to be padded u... — committed to tenstorrent/tt-metal by deleted user 2 years ago
- #12387: Workaround to_layout for height sharded tensor Conv2d will potentially trigger ttnn::to_layout for the input tensor in order to tilize the input. In cases where the input needs to be padded u... — committed to tenstorrent/tt-metal by deleted user 2 years ago
- #12387: Yolov4 sharded concat conv issue — committed to tenstorrent/tt-metal by keerthana-r-mcw 2 years ago
- #12387: Tweak UT to pass — committed to tenstorrent/tt-metal by deleted user 2 years ago
- #12387: Add concat UT which fails — committed to tenstorrent/tt-metal by deleted user 2 years ago
- #12387: Workaround to_layout for height sharded tensor Conv2d will potentially trigger ttnn::to_layout for the input tensor in order to tilize the input. In cases where the input needs to be padded u... — committed to tenstorrent/tt-metal by deleted user 2 years ago
- #12387: Workaround to_layout for height sharded tensor Conv2d will potentially trigger ttnn::to_layout for the input tensor in order to tilize the input. In cases where the input needs to be padded u... — committed to tenstorrent/tt-metal by deleted user 2 years ago
- #12387: Yolov4 sharded concat conv issue — committed to tenstorrent/tt-metal by keerthana-r-mcw 2 years ago
- #12387: Yolov4 sharded concat conv issue — committed to tenstorrent/tt-metal by keerthana-r-mcw 2 years ago
Please create a P0 issue for concat and assign to @tarafdarTT . We can then close this one.
I’ll merge the current PR I have for this, and for this concat issue we should open a new one and assign it to the concat op owner.