From bf3ab1362e5cdb00aa336660221176f0575fa494 Mon Sep 17 00:00:00 2001 From: vamsi-potturu Date: Wed, 21 Oct 2026 11:53:03 +0530 Subject: [PATCH 1/2] gen-ai: remove max_tokens from candidate inference config Co-authored-by: Cursor --- specification/resources/gen-ai/definitions.yml | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) diff --git a/specification/resources/gen-ai/definitions.yml b/specification/resources/gen-ai/definitions.yml index ce340c65a..8d49846d9 100644 --- a/specification/resources/gen-ai/definitions.yml +++ b/specification/resources/gen-ai/definitions.yml @@ -1534,12 +1534,11 @@ apiCandidateAgentConfig: type: string type: object apiCandidateInferenceConfig: - description: Inference configuration for the candidate model during evaluation. + description: |- + Inference configuration for the candidate model during evaluation. + The output token limit is not configurable. For serverless candidates it is + set from the model's max output tokens in the model catalog. properties: - max_tokens: - example: 123 - format: int64 - type: integer reasoning_effort: description: |- Reasoning effort for reasoning-capable models (e.g. "low", "medium", @@ -8165,7 +8164,7 @@ apiPresetSaveSection: - PRESET_SAVE_SECTION_CANDIDATE: Candidate model identity (`candidate_model_uuid`, `candidate_model_source`, `candidate_model_name`) and the non-prompt inference params - (`max_tokens`, `temperature`, `stop_token`). + (`temperature`, `stop_token`, `reasoning_effort`). - PRESET_SAVE_SECTION_METRICS: The selected `metric_uuids` and the optional `star_metric`. - PRESET_SAVE_SECTION_JUDGE: The `judge_model_uuid`. - PRESET_SAVE_SECTION_DATASET: The `dataset_uuid`. From 59abbfc0cab0b70f8bbbc061caa32efb878fad6d Mon Sep 17 00:00:00 2001 From: vamsi-potturu Date: Wed, 21 Oct 2026 12:38:22 +0530 Subject: [PATCH 2/2] gen-ai: simplify candidate inference config description Co-authored-by: Cursor --- specification/resources/gen-ai/definitions.yml | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/specification/resources/gen-ai/definitions.yml b/specification/resources/gen-ai/definitions.yml index 8d49846d9..298e720be 100644 --- a/specification/resources/gen-ai/definitions.yml +++ b/specification/resources/gen-ai/definitions.yml @@ -1536,8 +1536,7 @@ apiCandidateAgentConfig: apiCandidateInferenceConfig: description: |- Inference configuration for the candidate model during evaluation. - The output token limit is not configurable. For serverless candidates it is - set from the model's max output tokens in the model catalog. + The output token limit is set automatically and cannot be configured. properties: reasoning_effort: description: |-