Fix another bug in AIBench local bisector

Summary:
Fix the regression detector by importing the correct package name.
Deal with the case when the base commit metric could be zero.

Reviewed By: aaronenyeshi

Differential Revision: D51866054

fbshipit-source-id: 811dbdc688ce428bc000cebcf9402e651ca840b6

bisection.py

@@ -290,8 +290,9 @@ def get_digest_for_commit(self, commit: Commit, abtest_result: Dict[str, Any], d
         # Run benchmark, return the output json file
         result_json = self._run_benchmark_for_commit(commit, abtest_result)
         commit.digest = self._gen_digest(result_json)
-        print(f"================== [TORCHBENCH] Cleaning up packages for commit {commit.sha} ==================", flush=True)
-        cleanup_torch_packages()
+        if not IS_FBCODE:
+            print(f"================== [TORCHBENCH] Cleaning up packages for commit {commit.sha} ==================", flush=True)
+            cleanup_torch_packages()
         return commit.digest
 
 class TorchBenchBisection:

regression_detector.py

@@ -55,8 +55,11 @@ def _call_userbenchmark_detector(detector, control: Dict[str, Any], treatment: D
     assert control["name"] == treatment["name"], f'Expected the same userbenchmark name from metrics files, \
                                                 but getting {control["name"]} and {treatment["name"]}.'
     bm_name = control["name"]
-    detector = importlib.import_module(f"userbenchmark.{bm_name}.regression_detector").run
-
+    try:
+        detector = importlib.import_module(f"userbenchmark.{bm_name}.regression_detector").run
+    except:
+        # fbcode
+        detector = importlib.import_module(f"userbenchmark.fb.{bm_name}.regression_detector").run
     # Process control and treatment to include only shared keys
     filtered_control_metrics = {}
     control_only_metrics = {}
@@ -97,7 +100,7 @@ def process_regressions_into_yaml(regression_result: TorchBenchABTestResult, out
     with open(output_path, "w") as ofptr:
         ofptr.write(output_yaml_str)
     print(f"Wrote above yaml to {output_path}.")
-        
+
 
 def process_regressions_into_gh_issue(regression_result: TorchBenchABTestResult, owner: str, output_path: str, errors_path: str) -> None:
     regressions_dict = asdict(regression_result)
@@ -109,7 +112,7 @@ def process_regressions_into_gh_issue(regression_result: TorchBenchABTestResult,
             troubled_tests += f"- {test}: {sign}{delta:.5%}\n"
         else:
             troubled_tests += f"- {test}: {delta}\n"
-    
+
     control_only_tests = ""
     for test, stat in regressions_dict["control_only_metrics"].items():
         control_only_tests += f"- {test}: {stat}\n"
@@ -155,7 +158,7 @@ def process_regressions_into_gh_issue(regression_result: TorchBenchABTestResult,
         "github_run_url": github_run_url,
         "owner": owner
     }
-    
+
     issue_body = GITHUB_ISSUE_TEMPLATE.format(**issue_config)
     print(issue_body)
     with open(output_path, "w") as f: