Winograd Update for Small batches (#1803)

* Build in padding to Winograd kernels * Add new fused Winograd kernel * Enable weight flipping in Winograd kernels
2025-02-14 13:08:13 -08:00
parent 7aea5b1895
commit 2dc307f2e6
4 changed files with 505 additions and 86 deletions
@@ -3882,7 +3882,7 @@ array conv_general(

  return array(
      std::move(out_shape),
-      in.dtype(),
+      out_type,
      std::make_shared<Convolution>(
          to_stream(s),
          stride,