A video of 154 frames (batch_size=1, t=154) is flattend(bt=154, c=64, hw_sum=(11^2+22^2+44^2)) and input to Deformable Attention, but I encounted batch % im2colstep == 0 INTERNAL ASSERT FAILED at "/home/xuhuihui/workspace/Deformable-DETR/models/ops/src/cuda/ms_deform_attn_cuda.cu":52, please report a bug to PyTorch. batch(%d) must divide im2col_step(%d)15464.
In ms_deform_attn_cuda.cu, one line is const int im2colstep = std::min(batch, im2col_step);. Should I increase im2col_step to bigger than 154, like 256?
when i set imcol_step to 256, and no error encounted.
A video of 154 frames (batch_size=1, t=154) is flattend(bt=154, c=64, hw_sum=(11^2+22^2+44^2)) and input to Deformable Attention, but I encounted batch % im2colstep == 0 INTERNAL ASSERT FAILED at "/home/xuhuihui/workspace/Deformable-DETR/models/ops/src/cuda/ms_deform_attn_cuda.cu":52, please report a bug to PyTorch. batch(%d) must divide im2col_step(%d)15464.
In ms_deform_attn_cuda.cu, one line is const int im2colstep = std::min(batch, im2col_step);. Should I increase im2col_step to bigger than 154, like 256?
when i set imcol_step to 256, and no error encounted.