Update on "[aotinductor] Make writing of the weight files to be conditional"

desertfire · desertfire · commit f0c70337d338 · 2023-10-17T21:18:07.000-07:00
Summary: Since we cache the AOTInductor generated library file, we should not need to write the weights as binary file if the library file already exists.

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler

[ghstack-poisoned]
diff --git a/torch/_inductor/codecache.py b/torch/_inductor/codecache.py
@@ -1360,7 +1360,9 @@ def _to_bytes(t: torch.Tensor) -> bytes:
 
                         return bytes(raw_array.contents)
 
-                    aot_constants = b"".join(_to_bytes(tensor) for tensor in graph.constants.values())
+                    aot_constants = b"".join(
+                        _to_bytes(tensor) for tensor in graph.constants.values()
+                    )
 
                     consts_key, consts_path = write(
                         aot_constants,
