Convmixer (#1073)

* Only optimize float tensors.

* Use full tensors for zeros and ones.

* Add a benchmark for the matmul slowness.

* Add the convmixer model.

* Proper adaptive pooling.

candle-transformers/src/models/mod.rs

@@ -1,5 +1,6 @@
 pub mod bert;
 pub mod bigcode;
+pub mod convmixer;
 pub mod dinov2;
 pub mod efficientnet;
 pub mod falcon;