Add autotracing to eval step

Summary:
Pull Request resolved: #499

Add `prepare_fb_model_for_eval` override; no-op.

Reviewed By: frabu6

Differential Revision: D43906444

fbshipit-source-id: 97e06f1de8f3ba07808a0493d3d216031ff011d0

Author: metamate123

d2go/runner/default_runner.py

@@ -141,6 +141,11 @@ def prepare_fb_model(cfg: CfgNode, model: torch.nn.Module) -> torch.nn.Module:
     return model
 
 
+@fb_overwritable()
+def prepare_fb_model_for_eval(cfg: CfgNode, model: torch.nn.Module) -> torch.nn.Module:
+    return model
+
+
 class BaseRunner(object):
     def __init__(self):
         identifier = f"D2Go.Runner.{self.__class__.__name__}"
@@ -380,6 +385,7 @@ def _get_inference_dir_name(base_dir, inference_type, dataset_name):
             )
 
         attach_profilers(cfg, model)
+        prepare_fb_model_for_eval(cfg, model)
 
         results = OrderedDict()
         results[model_tag] = OrderedDict()