do not fuse model again for a QAT model

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/643

For a QAT model, it contains observers. After QAT training, those observers already contain updated statistics, such as min_val, max_val.

When we want to export FP32 QAT model for a sanity check, if we call **fuse_utils.fuse_model()** again (which is often already called when we build the QAT model before QAT training), it will remove statistics in the observers.

Reviewed By: wat3rBro

Differential Revision: D52152688

fbshipit-source-id: 08aa16f2aa72b3809e0ba2d346f1b806c0e6ede7

d2go/export/exporter.py

@@ -86,6 +86,8 @@ def _convert_fp_model(
     cfg: CfgNode, pytorch_model: nn.Module, data_loader: Iterable
 ) -> nn.Module:
     """Converts floating point predictor"""
+    if not isinstance(cfg, CfgNode) or (not cfg.QUANTIZATION.QAT.ENABLED):
+        # Do not fuse model again for QAT model since it will remove observer statistics (e.g. min_val, max_val)
         pytorch_model = fuse_utils.fuse_model(pytorch_model)
         logger.info(f"Fused Model:\n{pytorch_model}")
         if fuse_utils.count_bn_exist(pytorch_model) > 0: