fixing the checkpoint loading at inference-engine (#2429)

Co-authored-by: NAmmar Ahmad Awan <ammar.awan@microsoft.com>

deepspeed/inference/engine.py

@@ -449,7 +449,8 @@ class InferenceEngine(Module):
             ckpt_list = self._get_all_ckpt_names(load_dir, tag)
             sd_loader = SDLoaderFactory.get_sd_loader(ckpt_list, self.checkpoint_engine)
         else:
-            sd_loader = SDLoaderFactory.get_sd_loader_json(load_dir)
+            sd_loader = SDLoaderFactory.get_sd_loader_json(load_dir,
+                                                           self.checkpoint_engine)
 
         if type(sd_loader) is list:
             self.sd = torch.load(sd_loader[0], map_location='cpu')
@@ -492,7 +493,6 @@ class InferenceEngine(Module):
 
             self.module.load_state_dict(
                 state_dict=checkpoint[self._choose_module_key(checkpoint)],
-                checkpoint_engine=self.checkpoint_engine,
                 strict=load_module_strict)
 
     def _choose_module_key(self, sd):