add loggings for internal adoption tracking (#426)

Summary:
Pull Request resolved: #426

add adoption loggings for models with most internal usage. To avoid duplicated count, we only enable loggings for model entrypoints.
This diff includes: vision transformer, MLP, av_concat_fusion, contrastive loss and transformer fusion. CLIP logging has been enabled previously.

Reviewed By: ankitade

Differential Revision: D45532797

fbshipit-source-id: 7e1a2a56a99bc0fe180c1103402003dc1ad5cbe0

torchmultimodal/modules/layers/mlp.py

@@ -42,7 +42,7 @@ def __init__(
         normalization: Optional[Callable[..., nn.Module]] = None,
     ) -> None:
         super().__init__()
-
+        torch._C._log_api_usage_once(f"torchmultimodal.{self.__class__.__name__}")
         layers = nn.ModuleList()
 
         if hidden_dims is None:

torchmultimodal/modules/losses/contrastive_loss_with_temperature.py

@@ -165,6 +165,7 @@ def __init__(
         logit_scale_max: Optional[float] = math.log(100),
     ):
         super().__init__()
+        torch._C._log_api_usage_once(f"torchmultimodal.{self.__class__.__name__}")
 
         if not logit_scale_min and not logit_scale_max:
             raise ValueError(