allow to skip loading model weights in build_model()

Summary:

Currently, in runner **build_model()** method, when **eval_only=True**, we always try to load model weights.
This is quite restricted in some cases. For example, we may just wanna build a model in eval mode to profile its efficiency, and we have not trained the model or generated the model weights in a checkpoint file.

Thus, this diff adds an argument **skip_model_weights** to allow users to skip the loading of model weights.
Note, this diff is entirely back-compatible and is NOT expected to break existing implementations.

Differential Revision: D50623772

tools/exporter.py

@@ -45,6 +45,7 @@ def main(
     device: str = "cpu",
     compare_accuracy: bool = False,
     skip_if_fail: bool = False,
+    skip_model_weights: bool = False,
 ) -> ExporterOutput:
     if compare_accuracy:
         raise NotImplementedError(
@@ -58,8 +59,10 @@ def main(
 
     with temp_defrost(cfg):
         cfg.merge_from_list(["MODEL.DEVICE", device])
-    model = runner.build_model(cfg, eval_only=True)
+        if skip_model_weights:
+            cfg.merge_from_list(["MODEL.WEIGHTS", ""])
 
+    model = runner.build_model(cfg, eval_only=True)
     # NOTE: train dataset is used to avoid leakage since the data might be used for
     # running calibration for quantization. test_loader is used to make sure it follows
     # the inference behaviour (augmentation will not be applied).
@@ -112,6 +115,7 @@ def run_with_cmdline_args(args):
         device=args.device,
         compare_accuracy=args.compare_accuracy,
         skip_if_fail=args.skip_if_fail,
+        skip_model_weights=args.skip_model_weights,
     )
 
 
@@ -139,6 +143,8 @@ def get_parser():
         help="If set, suppress the exception for failed exporting and continue to"
         " export the next type of model",
     )
+    parser.add_argument("--skip-model-weights", action="store_true")
+
     return parser