Recently, I run others works with spconv library (spconv 2.3.6 cumm 0.4.11) and get an error:
<return ALL_TENSOR_OP_MAP[(shape_tuple, dtype_ab, dtype_c)]> KeyError: ((16, 8, 8), float, float)
After a while, I notice that, on cumm libaray we comment this line, link to code
# ((16, 8, 8), dtypes.float32, dtypes.float32): MmaM16N8K8F32((8, 0)),
Any reason why we comment this line? Even though when changed spconv layer to spconv.ConvAlgo.Native it works again but results seem to be a big difference.
Looking forward to your reply. Thanks for maintenance these two libraries.
Recently, I run others works with spconv library (spconv 2.3.6 cumm 0.4.11) and get an error:
After a while, I notice that, on cumm libaray we comment this line, link to code
Any reason why we comment this line? Even though when changed spconv layer to
spconv.ConvAlgo.Nativeit works again but results seem to be a big difference.Looking forward to your reply. Thanks for maintenance these two libraries.