Copy model before benchmark warmup runs (pytorch#145858)

Fixes pytorch#144772

The eager warmup runs causes the model to change state so that later when we export it, the model is different than when we export it directly out of box. For some reason exporting the model with the changed state causes issues but exporting the inital model is ok. This is the reason why the accuracy checks pass but the performance check fails when exporting.

Pull Request resolved: pytorch#145858
Approved by: https://github.com/desertfire

Author: angelayi

benchmarks/dynamo/common.py

@@ -3446,11 +3446,15 @@ def warmup(fn, model, example_inputs, mode, niters=5):
                 self.args.snapshot_memory, f"eager_{self.args.only}"
             ):
                 eager_latency, eager_peak_mem, _ = warmup(
-                    self.model_iter_fn, model, example_inputs, "eager"
+                    self.model_iter_fn, copy.deepcopy(model), example_inputs, "eager"
                 )
                 if self.args.use_warm_peak_memory:
                     _, eager_peak_mem, _ = warmup(
-                        self.model_iter_fn, model, example_inputs, "eager", niters=1
+                        self.model_iter_fn,
+                        copy.deepcopy(model),
+                        example_inputs,
+                        "eager",
+                        niters=1,
                     )
 
             if self.args.export_aot_inductor: