Closed sdjordjevicTT closed 1 week ago
We put a check in that requires the second input to have an explicit first dimension of size 1 in this scenario.
@TT-BrianLiu do you know how much work it'd be to allow the code to work in this scenario?
If it's too much work, we would need to update the documentation.
Batch dimensions don't line up regardless. Updated doc string with this scenario.
Doc string updated via PR https://github.com/tenstorrent/tt-metal/pull/13071
Describe the bug TTNN Matmul op does not work when the 3rd dimension of input_tensor_b isn't 1, hence the broadcast version of Matmul isn't executed. According to the public documentation, this kind of product is supported. I am consistently encountering an error when trying to execute this specific operation.
To Reproduce
Error once the above code is executed:
Expected behavior As stated in the public docs, in this particular case, the b input should be broadcasted from (7, 2048, 256) to (7, 7, 2048, 256) and the matrix product executed.
Screenshots N/A
Please complete the following environment information:
Additional context N/A