fix: unit tests

Author: mohammedahmed18

codeflash/result/critic.py

@@ -26,7 +26,10 @@ def performance_gain(*, original_runtime_ns: int, optimized_runtime_ns: int) ->
 
 
 def speedup_critic(
-    candidate_result: OptimizedCandidateResult, original_code_runtime: int, best_runtime_until_now: int
+    candidate_result: OptimizedCandidateResult,
+    original_code_runtime: int,
+    best_runtime_until_now: int,
+    disable_gh_action_noise: bool | None = None,
 ) -> bool:
     """Take in a correct optimized Test Result and decide if the optimization should actually be surfaced to the user.
 
@@ -35,10 +38,11 @@ def speedup_critic(
     when the original runtime is less than 10 microseconds, and becomes MIN_IMPROVEMENT_THRESHOLD for any higher runtime.
     The noise floor is doubled when benchmarking on a (noisy) GitHub Action virtual instance, also we want to be more confident there.
     """
-    in_github_actions_mode = bool(env_utils.get_pr_number())
     noise_floor = 3 * MIN_IMPROVEMENT_THRESHOLD if original_code_runtime < 10000 else MIN_IMPROVEMENT_THRESHOLD
-    if in_github_actions_mode:
-        noise_floor = noise_floor * 2  # Increase the noise floor in GitHub Actions mode
+    if not disable_gh_action_noise:
+        in_github_actions_mode = bool(env_utils.get_pr_number())
+        if in_github_actions_mode:
+            noise_floor = noise_floor * 2  # Increase the noise floor in GitHub Actions mode
 
     perf_gain = performance_gain(
         original_runtime_ns=original_code_runtime, optimized_runtime_ns=candidate_result.best_test_runtime

tests/test_critic.py

@@ -41,7 +41,7 @@ def test_speedup_critic() -> None:
         total_candidate_timing=12,
     )
 
-    assert speedup_critic(candidate_result, original_code_runtime, best_runtime_until_now)  # 20% improvement
+    assert speedup_critic(candidate_result, original_code_runtime, best_runtime_until_now, True)  # 20% improvement
 
     candidate_result = OptimizedCandidateResult(
         max_loop_count=5,
@@ -52,7 +52,7 @@ def test_speedup_critic() -> None:
         optimization_candidate_index=0,
     )
 
-    assert not speedup_critic(candidate_result, original_code_runtime, best_runtime_until_now)  # 6% improvement
+    assert not speedup_critic(candidate_result, original_code_runtime, best_runtime_until_now, True)  # 6% improvement
 
     original_code_runtime = 100000
     best_runtime_until_now = 100000
@@ -66,7 +66,7 @@ def test_speedup_critic() -> None:
         optimization_candidate_index=0,
     )
 
-    assert speedup_critic(candidate_result, original_code_runtime, best_runtime_until_now)  # 6% improvement
+    assert speedup_critic(candidate_result, original_code_runtime, best_runtime_until_now, True)  # 6% improvement
 
 
 def test_generated_test_critic() -> None: