4dec550e - Fixing test_linear to use a 3-D input tensor instead of 2-D one. The 2-D tensor was triggering eager mode bias fusion that impacted TF32 math on Ampere. (#1515)
Fixing test_linear to use a 3-D input tensor instead of 2-D one. The 2-D tensor was triggering eager mode bias fusion that impacted TF32 math on Ampere. (#1515)
