From add1ed2f0359fd5b4b58783c033dbff47e4aeca2 Mon Sep 17 00:00:00 2001 From: hrehard Date: Fri, 17 Jul 2026 16:35:47 -0700 Subject: [PATCH] fix: show LLM-as-judge results for Nova (InspectAI-routed) evaluations LLMAsJudgeEvaluator.show_results() raised '[PySDK Error] Could not extract training job name from pipeline steps' for Nova models even when the eval pipeline succeeded. Nova auto-routes to an InspectAI+Bedrock pipeline whose step is named 'EvaluateWithJudge', not EvaluateCustomModelMetrics / EvaluateBaseModelMetrics, so no job name was found. Fall back to that step; its training job holds the judge results in S3. --- .../src/sagemaker/train/common_utils/show_results_utils.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/sagemaker-train/src/sagemaker/train/common_utils/show_results_utils.py b/sagemaker-train/src/sagemaker/train/common_utils/show_results_utils.py index 55d478a730..02d56913fd 100644 --- a/sagemaker-train/src/sagemaker/train/common_utils/show_results_utils.py +++ b/sagemaker-train/src/sagemaker/train/common_utils/show_results_utils.py @@ -677,6 +677,11 @@ def _show_llmaj_results( pipeline_execution, "EvaluateBaseModelMetrics" ) + if not custom_job_name and not base_job_name: + custom_job_name = _extract_training_job_name_from_steps( + pipeline_execution, "EvaluateWithJudge" + ) + # Handle single-model evaluation scenarios if not custom_job_name and not base_job_name: raise ValueError(