set threshold lower

dphuang2 · dphuang2 · commit 58d81500e563 · 2025-08-05T15:45:07.000-07:00
diff --git a/tests/pytest/test_hallucination.py b/tests/pytest/test_hallucination.py
@@ -34,7 +34,7 @@ def hallucination_dataset_adapter(data: List[Dict[str, Any]]) -> List[Evaluation
     model=["accounts/fireworks/models/kimi-k2-instruct"],
     rollout_input_params=[{"temperature": 0.0, "max_tokens": 512}],
     rollout_processor=default_single_turn_rollout_processor,
-    threshold_of_success=1.0,
+    threshold_of_success=0.33,
     num_runs=1,
     mode="pointwise",
 )

Original file line number	Diff line number	Diff line change
`@@ -34,7 +34,7 @@ def hallucination_dataset_adapter(data: List[Dict[str, Any]]) -> List[Evaluation`
`34`	`34`	`model=["accounts/fireworks/models/kimi-k2-instruct"],`
`35`	`35`	`rollout_input_params=[{"temperature": 0.0, "max_tokens": 512}],`
`36`	`36`	`rollout_processor=default_single_turn_rollout_processor,`
`37`		`- threshold_of_success=1.0,`
	`37`	`+ threshold_of_success=0.33,`
`38`	`38`	`num_runs=1,`
`39`	`39`	`mode="pointwise",`
`40`	`40`	`)`